title stringlengths 2 74 ⌀ | summary stringlengths 11 1.09k ⌀ | github_url stringlengths 28 107 ⌀ | repo_title stringlengths 2 243 ⌀ | content_text stringlengths 0 277k ⌀ | content_html stringlengths 0 417k ⌀ | _source_url stringlengths 38 136 | _listing_url stringclasses 227
values |
|---|---|---|---|---|---|---|---|
Dash API Docs | MCP server for Dash, the macOS API documentation browser | https://github.com/Kapeli/dash-mcp-server | mcp-server-dash | mcp-server-dash A Model Context Protocol (MCP) server that provides tools to interact with the Dash documentation browser API. Dash 8 is required. You can download Dash 8 at https://blog.kapeli.com/dash-8 . <a href="https://glama.ai/mcp/servers/@Kapeli/dash-mcp-server"> <img width="380" height="200" src="https://glama.ai/mcp/servers/@Kapeli/dash-mcp-server/badge" alt="Dash Server MCP server" /> </a> Overview The Dash MCP server provides tools for accessing and searching documentation directly from Dash, the macOS documentation browser. MCP clients can: List installed docsets Search across docsets and code snippets Load documentation pages from search results Enable full-text search for specific docsets Notice This is a work in progress. Any suggestions are welcome! Tools list_installed_docsets Lists all installed documentation sets in Dash search_documentation Searches across docsets and snippets load_documentation_page Loads a documentation page from a load_url returned by search_documentation enable_docset_fts Enables full-text search for a specific docset Requirements macOS (required for Dash app) Dash installed Python 3.11.4 or higher uv Configuration Using uvx brew install uv in claude_desktop_config.json { "mcpServers": { "dash-api": { "command": "uvx", "args": [ "--from", "git+https://github.com/Kapeli/dash-mcp-server.git", "dash-mcp-server" ] } } } in Claude Code claude mcp add dash-api -- uvx --from "git+https://github.com/Kapeli/dash-mcp-server.git" "dash-mcp-server" | <h1>mcp-server-dash</h1>
<p>A Model Context Protocol (MCP) server that provides tools to interact with the <a href="https://kapeli.com/dash">Dash</a> documentation browser API.</p>
<p>Dash 8 is required. You can download Dash 8 at <a href="https://blog.kapeli.com/dash-8">https://blog.kapeli.com/dash-8</a>.</p>
<!-- --><a href="https://glama.ai/mcp/servers/@Kapeli/dash-mcp-server">
<img width="380" height="200" src="https://glama.ai/mcp/servers/@Kapeli/dash-mcp-server/badge" alt="Dash Server MCP server" />
</a><!-- -->
<h2>Overview</h2>
<p>The Dash MCP server provides tools for accessing and searching documentation directly from Dash, the macOS documentation browser. MCP clients can:</p>
<ul>
<li>List installed docsets</li>
<li>Search across docsets and code snippets</li>
<li>Load documentation pages from search results</li>
<li>Enable full-text search for specific docsets</li>
</ul>
<h3>Notice</h3>
<p>This is a work in progress. Any suggestions are welcome!</p>
<h2>Tools</h2>
<ol>
<li><strong>list_installed_docsets</strong>
<ul>
<li>Lists all installed documentation sets in Dash</li>
</ul>
</li>
<li><strong>search_documentation</strong>
<ul>
<li>Searches across docsets and snippets</li>
</ul>
</li>
<li><strong>load_documentation_page</strong>
<ul>
<li>Loads a documentation page from a <code>load_url</code> returned by <code>search_documentation</code></li>
</ul>
</li>
<li><strong>enable_docset_fts</strong>
<ul>
<li>Enables full-text search for a specific docset</li>
</ul>
</li>
</ol>
<h2>Requirements</h2>
<ul>
<li>macOS (required for Dash app)</li>
<li><a href="https://kapeli.com/dash">Dash</a> installed</li>
<li>Python 3.11.4 or higher</li>
<li>uv</li>
</ul>
<h2>Configuration</h2>
<h3>Using uvx</h3>
<pre><code class="language-bash">brew install uv
</code></pre>
<h4>in <code>claude_desktop_config.json</code></h4>
<pre><code class="language-json">{
"mcpServers": {
"dash-api": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/Kapeli/dash-mcp-server.git",
"dash-mcp-server"
]
}
}
}
</code></pre>
<h4>in <code>Claude Code</code></h4>
<pre><code class="language-bash">claude mcp add dash-api -- uvx --from "git+https://github.com/Kapeli/dash-mcp-server.git" "dash-mcp-server"
</code></pre> | https://mcpservers.org/servers/kapeli/dash-mcp-server | https://mcpservers.org/all?sort=newest&page=1 |
asterpay | EUR settlement for AI agents. USDC/EURC to EUR via SEPA Instant. Trust scoring (KYA 0-100), market data, crypto analytics, AI tools. | https://github.com/AsterPay/asterpay-mcp-server | null | No documentation available. | <p class="text-gray-500">No documentation available.</p> | https://mcpservers.org/servers/asterpay/asterpay-mcp-server | https://mcpservers.org/all?sort=newest&page=1 |
mcp-gmail | MCP server for full Gmail operations via Unipile API. 9 tools: send, reply, list, read, delete, search, labels, attachments, drafts. Dry-run by default, 55 unit tests. MIT licensed. | https://github.com/timkulbaev/mcp-gmail | null | No documentation available. | <p class="text-gray-500">No documentation available.</p> | https://mcpservers.org/servers/timkulbaev/mcp-gmail | https://mcpservers.org/all?sort=newest&page=1 |
mistaike.ai | MCP security gateway with DLP scanning (PII, secrets, API keys), prompt injection protection, Memory Vault, Bug Vault (295k+ patterns), and unified audit logging. Two endpoints: free bug search at /mcp and authenticated hub at /hub_mcp. | https://github.com/mistaike-ai/mistaike-mcp | null | No documentation available. | <p class="text-gray-500">No documentation available.</p> | https://mcpservers.org/servers/mistaike-ai/mistaike-mcp | https://mcpservers.org/all?sort=newest&page=1 |
Swap API | Free token swaps for AI agents. No API keys. Returns executable transaction calldata for 40+ EVM chains. | https://github.com/Swap-API/swap-api | Swap API | Swap API Executable token swap calldata in one GET request. No API keys. No accounts. No SDK bloat. https://api.swapapi.dev Quick Start Swap 1 ETH for USDC on Ethereum: curl "https://api.swapapi.dev/v1/swap/1?\ tokenIn=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE&\ tokenOut=0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48&\ amount=1000000000000000000&\ sender=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045" That's it. The response contains everything you need to sign and broadcast. Example Response { "success": true, "data": { "status": "Successful", "tokenFrom": { "address": "0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE", "symbol": "ETH", "name": "Ether", "decimals": 18 }, "tokenTo": { "address": "0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48", "symbol": "USDC", "name": "USD Coin", "decimals": 6 }, "swapPrice": 2435.12, "priceImpact": 0.0003, "amountIn": "1000000000000000000", "expectedAmountOut": "2435120000", "minAmountOut": "2422947280", "tx": { "from": "0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045", "to": "0x011E52E4E40CF9498c79e329EBc29ed08c8B5abB", "data": "0x2646478b...", "value": "1000000000000000000", "gasPrice": 30000000000, "gas": "250000" } }, "timestamp": "2026-03-12T00:00:00.000Z" } Field Description success Boolean indicating request success data.status "Successful" , "Partial" , or "NoRoute" data.tokenFrom/tokenTo Token metadata (address, symbol, decimals) data.swapPrice Exchange rate data.priceImpact Slippage impact (0.001 = 0.1%) data.expectedAmountOut Estimated output in token's smallest unit data.minAmountOut Guaranteed minimum (respects your maxSlippage ) data.tx Transaction object ready to sign and send API Reference Endpoint: GET /v1/swap/{chainId} Parameters: Param Required Description chainId path Chain ID (1=Ethereum, 8453=Base, 42161=Arbitrum, etc.) tokenIn query Input token address. Use 0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE for native ETH tokenOut query Output token address amount query Input amount in smallest unit (e.g., wei for ETH) sender query Your wallet address (used to build the tx) maxSlippage query Optional. 0-1 (default: 0.005 = 0.5%) Response codes: 200 — Quote ready 400 — Invalid params or unsupported chain 429 — Rate limit exceeded (60/min per IP) 502 — Upstream service error More Curl Examples Base (ETH → USDC) curl "https://api.swapapi.dev/v1/swap/8453?\ tokenIn=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE&\ tokenOut=0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913&\ amount=500000000000000000&\ sender=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045" Arbitrum (USDC → ETH with 1% slippage) curl "https://api.swapapi.dev/v1/swap/42161?\ tokenIn=0xaf88d065e77c8cC2239327C5EDb3A432268e5831&\ tokenOut=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE&\ amount=1000000000&\ sender=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045&\ maxSlippage=0.01" Ethereum Mainnet (DAI → USDC) curl "https://api.swapapi.dev/v1/swap/1?\ tokenIn=0x6B175474E89094C44Da98b954EedeAC495271d0F&\ tokenOut=0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48&\ amount=1000000000000000000000&\ sender=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045" Polygon (MATIC → USDC) curl "https://api.swapapi.dev/v1/swap/137?\ tokenIn=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE&\ tokenOut=0x2791Bca1f2de4661ED88A30C99A7a9449Aa84174&\ amount=10000000000000000000&\ sender=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045" Executing Swaps The API gives you an unsigned transaction. Sign it and broadcast: With Foundry cast # 1. Set swap parameters CHAIN_ID=8453 TOKEN_IN=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE TOKEN_OUT=0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913 AMOUNT=1000000000000000000 SENDER=0x... # Your address PRIVATE_KEY=0x... # Your private key RPC_URL=https://mainnet.base.org # 2. Get swap quote RESPONSE=$(curl -s "https://api.swapapi.dev/v1/swap/$CHAIN_ID?\ tokenIn=$TOKEN_IN&\ tokenOut=$TOKEN_OUT&\ amount=$AMOUNT&\ sender=$SENDER") # 3. Parse the tx fields TX_TO=$(echo "$RESPONSE" | jq -r '.data.tx.to') TX_DATA=$(echo "$RESPONSE" | jq -r '.data.tx.data') TX_VALUE=$(echo "$RESPONSE" | jq -r '.data.tx.value') TX_GAS=$(echo "$RESPONSE" | jq -r '.data.tx.gas') # 4. Sign and send cast send \ --rpc-url "$RPC_URL" \ --private-key "$PRIVATE_KEY" \ "$TX_TO" \ --value "$TX_VALUE" \ --gas-limit "$TX_GAS" \ --data "$TX_DATA" With viem import { createWalletClient, http } from 'viem' import { privateKeyToAccount } from 'viem/accounts' import { base } from 'viem/chains' const account = privateKeyToAccount('0x...') const client = createWalletClient({ account, chain: base, transport: http() }) const response = await fetch( 'https://api.swapapi.dev/v1/swap/8453?' + 'tokenIn=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE&' + 'tokenOut=0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913&' + 'amount=1000000000000000000&' + 'sender=' + account.address ) const { data } = await response.json() const hash = await client.sendTransaction({ to: data.tx.to, data: data.tx.data, value: BigInt(data.tx.value), gas: BigInt(data.tx.gas) }) await client.waitForTransactionReceipt({ hash }) MCP Server For AI agents, Claude Desktop, Cursor, Cline, and other MCP clients: npx @swapapi/mcp Or add to Claude Desktop config: { "mcpServers": { "swapapi": { "command": "npx", "args": ["@swapapi/mcp"] } } } See mcp/README.md for full MCP documentation. OpenAPI Spec See openapi.json for the full OpenAPI specification. Limits Rate limit: 60 requests/minute per IP No authentication required Free to use License MIT | <h1>Swap API</h1>
<p><strong>Executable token swap calldata in one GET request.</strong> No API keys. No accounts. No SDK bloat.</p>
<pre><code>https://api.swapapi.dev
</code></pre>
<hr>
<h2>Quick Start</h2>
<p>Swap 1 ETH for USDC on Ethereum:</p>
<pre><code class="language-bash">curl "https://api.swapapi.dev/v1/swap/1?\
tokenIn=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE&\
tokenOut=0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48&\
amount=1000000000000000000&\
sender=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045"
</code></pre>
<p>That's it. The response contains everything you need to sign and broadcast.</p>
<hr>
<h2>Example Response</h2>
<pre><code class="language-json">{
"success": true,
"data": {
"status": "Successful",
"tokenFrom": {
"address": "0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE",
"symbol": "ETH",
"name": "Ether",
"decimals": 18
},
"tokenTo": {
"address": "0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48",
"symbol": "USDC",
"name": "USD Coin",
"decimals": 6
},
"swapPrice": 2435.12,
"priceImpact": 0.0003,
"amountIn": "1000000000000000000",
"expectedAmountOut": "2435120000",
"minAmountOut": "2422947280",
"tx": {
"from": "0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045",
"to": "0x011E52E4E40CF9498c79e329EBc29ed08c8B5abB",
"data": "0x2646478b...",
"value": "1000000000000000000",
"gasPrice": 30000000000,
"gas": "250000"
}
},
"timestamp": "2026-03-12T00:00:00.000Z"
}
</code></pre>
<table><thead><tr><th>Field</th><th>Description</th></tr></thead><tbody><tr><td><code>success</code></td><td>Boolean indicating request success</td></tr><tr><td><code>data.status</code></td><td><code>"Successful"</code>, <code>"Partial"</code>, or <code>"NoRoute"</code></td></tr><tr><td><code>data.tokenFrom/tokenTo</code></td><td>Token metadata (address, symbol, decimals)</td></tr><tr><td><code>data.swapPrice</code></td><td>Exchange rate</td></tr><tr><td><code>data.priceImpact</code></td><td>Slippage impact (0.001 = 0.1%)</td></tr><tr><td><code>data.expectedAmountOut</code></td><td>Estimated output in token's smallest unit</td></tr><tr><td><code>data.minAmountOut</code></td><td>Guaranteed minimum (respects your <code>maxSlippage</code>)</td></tr><tr><td><code>data.tx</code></td><td>Transaction object ready to sign and send</td></tr></tbody></table>
<hr>
<h2>API Reference</h2>
<p><strong>Endpoint:</strong> <code>GET /v1/swap/{chainId}</code></p>
<p><strong>Parameters:</strong></p>
<table><thead><tr><th>Param</th><th>Required</th><th>Description</th></tr></thead><tbody><tr><td><code>chainId</code></td><td>path</td><td>Chain ID (1=Ethereum, 8453=Base, 42161=Arbitrum, etc.)</td></tr><tr><td><code>tokenIn</code></td><td>query</td><td>Input token address. Use <code>0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE</code> for native ETH</td></tr><tr><td><code>tokenOut</code></td><td>query</td><td>Output token address</td></tr><tr><td><code>amount</code></td><td>query</td><td>Input amount in smallest unit (e.g., wei for ETH)</td></tr><tr><td><code>sender</code></td><td>query</td><td>Your wallet address (used to build the tx)</td></tr><tr><td><code>maxSlippage</code></td><td>query</td><td>Optional. 0-1 (default: 0.005 = 0.5%)</td></tr></tbody></table>
<p><strong>Response codes:</strong></p>
<ul>
<li><code>200</code> — Quote ready</li>
<li><code>400</code> — Invalid params or unsupported chain</li>
<li><code>429</code> — Rate limit exceeded (60/min per IP)</li>
<li><code>502</code> — Upstream service error</li>
</ul>
<hr>
<h2>More Curl Examples</h2>
<h3>Base (ETH → USDC)</h3>
<pre><code class="language-bash">curl "https://api.swapapi.dev/v1/swap/8453?\
tokenIn=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE&\
tokenOut=0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913&\
amount=500000000000000000&\
sender=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045"
</code></pre>
<h3>Arbitrum (USDC → ETH with 1% slippage)</h3>
<pre><code class="language-bash">curl "https://api.swapapi.dev/v1/swap/42161?\
tokenIn=0xaf88d065e77c8cC2239327C5EDb3A432268e5831&\
tokenOut=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE&\
amount=1000000000&\
sender=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045&\
maxSlippage=0.01"
</code></pre>
<h3>Ethereum Mainnet (DAI → USDC)</h3>
<pre><code class="language-bash">curl "https://api.swapapi.dev/v1/swap/1?\
tokenIn=0x6B175474E89094C44Da98b954EedeAC495271d0F&\
tokenOut=0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48&\
amount=1000000000000000000000&\
sender=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045"
</code></pre>
<h3>Polygon (MATIC → USDC)</h3>
<pre><code class="language-bash">curl "https://api.swapapi.dev/v1/swap/137?\
tokenIn=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE&\
tokenOut=0x2791Bca1f2de4661ED88A30C99A7a9449Aa84174&\
amount=10000000000000000000&\
sender=0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045"
</code></pre>
<hr>
<h2>Executing Swaps</h2>
<p>The API gives you an unsigned transaction. Sign it and broadcast:</p>
<h3>With Foundry <code>cast</code></h3>
<pre><code class="language-bash"># 1. Set swap parameters
CHAIN_ID=8453
TOKEN_IN=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE
TOKEN_OUT=0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913
AMOUNT=1000000000000000000
SENDER=0x... # Your address
PRIVATE_KEY=0x... # Your private key
RPC_URL=https://mainnet.base.org
# 2. Get swap quote
RESPONSE=$(curl -s "https://api.swapapi.dev/v1/swap/$CHAIN_ID?\
tokenIn=$TOKEN_IN&\
tokenOut=$TOKEN_OUT&\
amount=$AMOUNT&\
sender=$SENDER")
# 3. Parse the tx fields
TX_TO=$(echo "$RESPONSE" | jq -r '.data.tx.to')
TX_DATA=$(echo "$RESPONSE" | jq -r '.data.tx.data')
TX_VALUE=$(echo "$RESPONSE" | jq -r '.data.tx.value')
TX_GAS=$(echo "$RESPONSE" | jq -r '.data.tx.gas')
# 4. Sign and send
cast send \
--rpc-url "$RPC_URL" \
--private-key "$PRIVATE_KEY" \
"$TX_TO" \
--value "$TX_VALUE" \
--gas-limit "$TX_GAS" \
--data "$TX_DATA"
</code></pre>
<h3>With viem</h3>
<pre><code class="language-typescript">import { createWalletClient, http } from 'viem'
import { privateKeyToAccount } from 'viem/accounts'
import { base } from 'viem/chains'
const account = privateKeyToAccount('0x...')
const client = createWalletClient({
account,
chain: base,
transport: http()
})
const response = await fetch(
'https://api.swapapi.dev/v1/swap/8453?' +
'tokenIn=0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE&' +
'tokenOut=0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913&' +
'amount=1000000000000000000&' +
'sender=' + account.address
)
const { data } = await response.json()
const hash = await client.sendTransaction({
to: data.tx.to,
data: data.tx.data,
value: BigInt(data.tx.value),
gas: BigInt(data.tx.gas)
})
await client.waitForTransactionReceipt({ hash })
</code></pre>
<h2>MCP Server</h2>
<p>For AI agents, Claude Desktop, Cursor, Cline, and other MCP clients:</p>
<pre><code class="language-bash">npx @swapapi/mcp
</code></pre>
<p>Or add to Claude Desktop config:</p>
<pre><code class="language-json">{
"mcpServers": {
"swapapi": {
"command": "npx",
"args": ["@swapapi/mcp"]
}
}
}
</code></pre>
<p>See <a href="./mcp/README.md">mcp/README.md</a> for full MCP documentation.</p>
<hr>
<h2>OpenAPI Spec</h2>
<p>See <a href="./openapi.json">openapi.json</a> for the full OpenAPI specification.</p>
<hr>
<h2>Limits</h2>
<ul>
<li><strong>Rate limit:</strong> 60 requests/minute per IP</li>
<li><strong>No authentication required</strong></li>
<li><strong>Free to use</strong></li>
</ul>
<hr>
<h2>License</h2>
<p>MIT</p> | https://mcpservers.org/servers/swap-api/swap-api | https://mcpservers.org/all?sort=newest&page=1 |
Timergy MCP Server | Create scheduling polls and find the perfect meeting time. No auth required. | https://github.com/timergy-app/timergy | @timergy/mcp | @timergy/mcp MCP server for Timergy - the scheduling poll app. Create polls, vote on time slots, and finalize meetings directly from AI agents like Claude, ChatGPT, and Gemini. No authentication required. Works out of the box. What is Timergy? Timergy is a scheduling poll service (like Doodle or When2Meet). You create a poll with time slot options, share it with participants, everyone votes on their availability, and you pick the best time. Quick Start Claude Desktop / Claude Code Add to your MCP config ( ~/.claude/claude_desktop_config.json or project .mcp.json ): { "mcpServers": { "timergy": { "command": "npx", "args": ["@timergy/mcp"] } } } Then just ask: "Create a poll for dinner next Friday or Saturday evening" Smithery (Remote, no install) smithery mcp add timergy/timergy Or connect directly via https://timergy--timergy.run.tools Other MCP Clients Any MCP-compatible client can use this server. Supports stdio (local) and Streamable HTTP (remote at https://api.timergy.com/mcp ). Tools create_poll Create a scheduling poll with time slot options. Parameter Type Required Description title string yes Poll title options array yes Time slots ( {start, end} in ISO 8601 with timezone) description string no Poll description deadline string no Voting deadline (ISO 8601) location string no Event location creatorName string no Name shown as poll creator "Create a poll called 'Team Lunch' with slots on Monday 12-13 and Tuesday 12-13" Returns: poll URL (to share), passphrase (for admin access), option IDs. get_poll Get poll details and time slot options (no votes). Use this to retrieve option IDs before voting or finalizing. Parameter Type Required Description pollId string yes Poll UUID "Show me the details of poll abc-123" vote_on_poll Submit votes on a poll. Each vote maps a time slot to yes/maybe/no. Parameter Type Required Description pollId string yes Poll UUID voterName string yes Name of the voter voterEmail string no Voter email (for notifications) votes array yes {optionId, availability} where availability is yes / maybe / no "Vote 'yes' for Monday and 'no' for Tuesday on poll abc-123 as Max" get_results See who voted and which time slots are most popular. Parameter Type Required Description pollId string yes Poll UUID "Show me the results for poll abc-123" finalize_poll Lock in the winning time slot. Notifies participants who provided an email. Parameter Type Required Description pollId string yes Poll UUID optionId string yes Winning time slot UUID passphrase string no Admin passphrase (auto-remembered from create_poll) "Finalize poll abc-123 with the Monday slot" Workflow 1. create_poll -> Get poll URL + passphrase 2. Share URL -> Send to participants 3. get_results -> See who voted yes/maybe/no 4. finalize_poll -> Pick the best time, lock it in The passphrase from step 1 is automatically remembered for step 4 within the same session. REST API The MCP server calls the Timergy Open API under the hood. You can also use the API directly: Method Endpoint Description POST /api/open/polls Create a poll GET /api/open/polls/:id Get poll details GET /api/open/polls/:id/options Get time slots GET /api/open/polls/:id/results Get vote results POST /api/open/polls/:id/vote Submit votes POST /api/open/polls/:id/admin-token Get admin token POST /api/open/polls/:id/finalize Finalize poll Base URL: https://api.timergy.com OpenAPI spec: https://api.timergy.com/api/open/openapi.json Configuration Environment Variable Default Description TIMERGY_API_URL https://api.timergy.com API base URL For local development: { "mcpServers": { "timergy": { "command": "npx", "args": ["@timergy/mcp"], "env": { "TIMERGY_API_URL": "http://localhost:4000" } } } } Rate Limits Action Limit Create poll 5 per hour Vote 20 per minute Get poll/results 60 per minute Admin token 5 per 10 minutes Finalize 3 per hour License MIT | <h1>@timergy/mcp</h1>
<p><a href="https://smithery.ai/server/timergy/timergy"><img alt="Smithery" src="https://smithery.ai/badge/timergy/timergy"></a></p>
<p>MCP server for <a href="https://timergy.com">Timergy</a> - the scheduling poll app. Create polls, vote on time slots, and finalize meetings directly from AI agents like Claude, ChatGPT, and Gemini.</p>
<p><strong>No authentication required.</strong> Works out of the box.</p>
<h2>What is Timergy?</h2>
<p>Timergy is a scheduling poll service (like Doodle or When2Meet). You create a poll with time slot options, share it with participants, everyone votes on their availability, and you pick the best time.</p>
<h2>Quick Start</h2>
<h3>Claude Desktop / Claude Code</h3>
<p>Add to your MCP config (<code>~/.claude/claude_desktop_config.json</code> or project <code>.mcp.json</code>):</p>
<pre><code class="language-json">{
"mcpServers": {
"timergy": {
"command": "npx",
"args": ["@timergy/mcp"]
}
}
}
</code></pre>
<p>Then just ask:</p>
<blockquote>
<p>"Create a poll for dinner next Friday or Saturday evening"</p>
</blockquote>
<h3>Smithery (Remote, no install)</h3>
<pre><code>smithery mcp add timergy/timergy
</code></pre>
<p>Or connect directly via <code>https://timergy--timergy.run.tools</code></p>
<h3>Other MCP Clients</h3>
<p>Any MCP-compatible client can use this server. Supports stdio (local) and Streamable HTTP (remote at <code>https://api.timergy.com/mcp</code>).</p>
<h2>Tools</h2>
<h3><code>create_poll</code></h3>
<p>Create a scheduling poll with time slot options.</p>
<table><thead><tr><th>Parameter</th><th>Type</th><th>Required</th><th>Description</th></tr></thead><tbody><tr><td><code>title</code></td><td>string</td><td>yes</td><td>Poll title</td></tr><tr><td><code>options</code></td><td>array</td><td>yes</td><td>Time slots (<code>{start, end}</code> in ISO 8601 with timezone)</td></tr><tr><td><code>description</code></td><td>string</td><td>no</td><td>Poll description</td></tr><tr><td><code>deadline</code></td><td>string</td><td>no</td><td>Voting deadline (ISO 8601)</td></tr><tr><td><code>location</code></td><td>string</td><td>no</td><td>Event location</td></tr><tr><td><code>creatorName</code></td><td>string</td><td>no</td><td>Name shown as poll creator</td></tr></tbody></table>
<pre><code>"Create a poll called 'Team Lunch' with slots on Monday 12-13 and Tuesday 12-13"
</code></pre>
<p>Returns: poll URL (to share), passphrase (for admin access), option IDs.</p>
<h3><code>get_poll</code></h3>
<p>Get poll details and time slot options (no votes). Use this to retrieve option IDs before voting or finalizing.</p>
<table><thead><tr><th>Parameter</th><th>Type</th><th>Required</th><th>Description</th></tr></thead><tbody><tr><td><code>pollId</code></td><td>string</td><td>yes</td><td>Poll UUID</td></tr></tbody></table>
<pre><code>"Show me the details of poll abc-123"
</code></pre>
<h3><code>vote_on_poll</code></h3>
<p>Submit votes on a poll. Each vote maps a time slot to yes/maybe/no.</p>
<table><thead><tr><th>Parameter</th><th>Type</th><th>Required</th><th>Description</th></tr></thead><tbody><tr><td><code>pollId</code></td><td>string</td><td>yes</td><td>Poll UUID</td></tr><tr><td><code>voterName</code></td><td>string</td><td>yes</td><td>Name of the voter</td></tr><tr><td><code>voterEmail</code></td><td>string</td><td>no</td><td>Voter email (for notifications)</td></tr><tr><td><code>votes</code></td><td>array</td><td>yes</td><td><code>{optionId, availability}</code> where availability is <code>yes</code>/<code>maybe</code>/<code>no</code></td></tr></tbody></table>
<pre><code>"Vote 'yes' for Monday and 'no' for Tuesday on poll abc-123 as Max"
</code></pre>
<h3><code>get_results</code></h3>
<p>See who voted and which time slots are most popular.</p>
<table><thead><tr><th>Parameter</th><th>Type</th><th>Required</th><th>Description</th></tr></thead><tbody><tr><td><code>pollId</code></td><td>string</td><td>yes</td><td>Poll UUID</td></tr></tbody></table>
<pre><code>"Show me the results for poll abc-123"
</code></pre>
<h3><code>finalize_poll</code></h3>
<p>Lock in the winning time slot. Notifies participants who provided an email.</p>
<table><thead><tr><th>Parameter</th><th>Type</th><th>Required</th><th>Description</th></tr></thead><tbody><tr><td><code>pollId</code></td><td>string</td><td>yes</td><td>Poll UUID</td></tr><tr><td><code>optionId</code></td><td>string</td><td>yes</td><td>Winning time slot UUID</td></tr><tr><td><code>passphrase</code></td><td>string</td><td>no</td><td>Admin passphrase (auto-remembered from create_poll)</td></tr></tbody></table>
<pre><code>"Finalize poll abc-123 with the Monday slot"
</code></pre>
<h2>Workflow</h2>
<pre><code>1. create_poll -> Get poll URL + passphrase
2. Share URL -> Send to participants
3. get_results -> See who voted yes/maybe/no
4. finalize_poll -> Pick the best time, lock it in
</code></pre>
<p>The passphrase from step 1 is automatically remembered for step 4 within the same session.</p>
<h2>REST API</h2>
<p>The MCP server calls the Timergy Open API under the hood. You can also use the API directly:</p>
<table><thead><tr><th>Method</th><th>Endpoint</th><th>Description</th></tr></thead><tbody><tr><td><code>POST</code></td><td><code>/api/open/polls</code></td><td>Create a poll</td></tr><tr><td><code>GET</code></td><td><code>/api/open/polls/:id</code></td><td>Get poll details</td></tr><tr><td><code>GET</code></td><td><code>/api/open/polls/:id/options</code></td><td>Get time slots</td></tr><tr><td><code>GET</code></td><td><code>/api/open/polls/:id/results</code></td><td>Get vote results</td></tr><tr><td><code>POST</code></td><td><code>/api/open/polls/:id/vote</code></td><td>Submit votes</td></tr><tr><td><code>POST</code></td><td><code>/api/open/polls/:id/admin-token</code></td><td>Get admin token</td></tr><tr><td><code>POST</code></td><td><code>/api/open/polls/:id/finalize</code></td><td>Finalize poll</td></tr></tbody></table>
<p>Base URL: <code>https://api.timergy.com</code></p>
<p>OpenAPI spec: <code>https://api.timergy.com/api/open/openapi.json</code></p>
<h2>Configuration</h2>
<table><thead><tr><th>Environment Variable</th><th>Default</th><th>Description</th></tr></thead><tbody><tr><td><code>TIMERGY_API_URL</code></td><td><code>https://api.timergy.com</code></td><td>API base URL</td></tr></tbody></table>
<p>For local development:</p>
<pre><code class="language-json">{
"mcpServers": {
"timergy": {
"command": "npx",
"args": ["@timergy/mcp"],
"env": { "TIMERGY_API_URL": "http://localhost:4000" }
}
}
}
</code></pre>
<h2>Rate Limits</h2>
<table><thead><tr><th>Action</th><th>Limit</th></tr></thead><tbody><tr><td>Create poll</td><td>5 per hour</td></tr><tr><td>Vote</td><td>20 per minute</td></tr><tr><td>Get poll/results</td><td>60 per minute</td></tr><tr><td>Admin token</td><td>5 per 10 minutes</td></tr><tr><td>Finalize</td><td>3 per hour</td></tr></tbody></table>
<h2>License</h2>
<p>MIT</p> | https://mcpservers.org/servers/timergy-app/timergy | https://mcpservers.org/all?sort=newest&page=1 |
Skillbase/spm | npm for AI skills. Create, share, and discover reusable AI instructions for any model via MCP. | https://github.com/useskillbase/spm | Skillbase (spm) | Skillbase (spm) AI skills manager — install, publish, and deploy reusable prompts, personas, and MCP tools across 14 AI clients. Think npm for AI capabilities : versioned packages of prompts, instructions, and tools that any LLM can use. npm install -g @skillbase/spm Why Skillbase? One command to connect — spm connect claude wires up MCP for Claude, Cursor, VS Code, Zed, and 10 more clients Portable skills — write once, use everywhere. Skills are model-agnostic packages with semver, dependencies, and a registry Personas — define AI agent personalities with traits, model settings, and skill dependencies in a single .person.json Deploy targets — export and deploy personas to external platforms (OpenClaw, more coming) Built-in MCP server — skills auto-load into your AI client via Model Context Protocol Supported AI Clients Client Connect command Claude Desktop spm connect claude Claude Code spm connect claude-code Cursor spm connect cursor VS Code (Copilot) spm connect vscode Windsurf spm connect windsurf Zed spm connect zed JetBrains IDEs spm connect jetbrains Cline spm connect cline Roo Code spm connect roo-code Continue spm connect continue Amazon Q Developer spm connect amazonq Gemini CLI spm connect gemini OpenCode spm connect opencode OpenClaw spm connect openclaw Quick Start # Initialize skills directory spm init # Search for skills spm search "code review" # Install a skill spm add author/skill-name # Connect to your AI client spm connect claude What is a Skill? A skill is a portable, versioned package containing prompts, tools, or instructions that any AI model can use. my-skill/ ├── skill.json # Manifest (name, version, triggers, dependencies) └── SKILL.md # Main prompt / instructions Personas Define AI agent personalities with character traits, model settings, and skill dependencies: # Create a persona spm persona create my-agent # Activate (auto-installs missing skills) spm persona activate my-agent # Export to external platform spm persona export my-agent -f openclaw # Deploy spm persona deploy my-agent -t openclaw Commands Manage Skills Command Description spm add <ref> Install a skill spm install Install all dependencies from skill.json spm remove <ref> Remove a skill spm create <name> Scaffold a new skill spm link <path> Symlink a local skill for development spm convert <file> Convert .md/.txt prompts into skills spm list List installed skills spm info <name> Show skill details spm validate Validate a skill directory Personas Command Description spm persona create Create a new persona spm persona list List installed personas spm persona activate Activate persona (auto-installs skills) spm persona deactivate Deactivate current persona spm persona export Export to target platform format spm persona deploy Deploy to target platform spm persona import Import from external platform Registry Command Description spm search <query> Search local and remote registries spm publish <path> Publish to registry spm update <path> Update a published skill spm login Authenticate (GitHub OAuth) spm rate <name> Rate a skill (1-5) spm registry add <url> Add a remote registry System Command Description spm connect <client> Connect MCP server to AI client spm disconnect <client> Disconnect from AI client spm serve Start MCP server (stdio) spm init Initialize skills directory spm reindex Rebuild skill index MCP Server Skillbase includes a built-in Model Context Protocol server that exposes installed skills as tools to any MCP-compatible AI client: spm serve --stdio MCP tools provided: skill_list , skill_load , skill_search , skill_context , skill_feedback , skill_install , persona_list , persona_load Registries Skills can be published to self-hosted registries or installed directly from GitHub: # Install from registry spm add author/skill-name # Install from GitHub spm add github:author/repo # Add a custom registry spm registry add https://registry.example.com # Publish spm publish ./my-skill Requirements Node.js >= 20.0.0 License MIT | <h1>Skillbase (spm)</h1>
<p><a href="https://www.npmjs.com/package/@skillbase/spm"><img alt="npm version" src="https://img.shields.io/npm/v/@skillbase/spm"></a>
<a href="https://github.com/useskillbase/spm/blob/main/LICENSE"><img alt="license" src="https://img.shields.io/npm/l/@skillbase/spm"></a>
<a href="https://www.npmjs.com/package/@skillbase/spm"><img alt="downloads" src="https://img.shields.io/npm/dm/@skillbase/spm"></a></p>
<p><strong>AI skills manager</strong> — install, publish, and deploy reusable prompts, personas, and MCP tools across 14 AI clients.</p>
<p>Think <strong>npm for AI capabilities</strong>: versioned packages of prompts, instructions, and tools that any LLM can use.</p>
<pre><code class="language-bash">npm install -g @skillbase/spm
</code></pre>
<h2>Why Skillbase?</h2>
<ul>
<li><strong>One command to connect</strong> — <code>spm connect claude</code> wires up MCP for Claude, Cursor, VS Code, Zed, and 10 more clients</li>
<li><strong>Portable skills</strong> — write once, use everywhere. Skills are model-agnostic packages with semver, dependencies, and a registry</li>
<li><strong>Personas</strong> — define AI agent personalities with traits, model settings, and skill dependencies in a single <code>.person.json</code></li>
<li><strong>Deploy targets</strong> — export and deploy personas to external platforms (OpenClaw, more coming)</li>
<li><strong>Built-in MCP server</strong> — skills auto-load into your AI client via Model Context Protocol</li>
</ul>
<h2>Supported AI Clients</h2>
<table><thead><tr><th>Client</th><th>Connect command</th></tr></thead><tbody><tr><td>Claude Desktop</td><td><code>spm connect claude</code></td></tr><tr><td>Claude Code</td><td><code>spm connect claude-code</code></td></tr><tr><td>Cursor</td><td><code>spm connect cursor</code></td></tr><tr><td>VS Code (Copilot)</td><td><code>spm connect vscode</code></td></tr><tr><td>Windsurf</td><td><code>spm connect windsurf</code></td></tr><tr><td>Zed</td><td><code>spm connect zed</code></td></tr><tr><td>JetBrains IDEs</td><td><code>spm connect jetbrains</code></td></tr><tr><td>Cline</td><td><code>spm connect cline</code></td></tr><tr><td>Roo Code</td><td><code>spm connect roo-code</code></td></tr><tr><td>Continue</td><td><code>spm connect continue</code></td></tr><tr><td>Amazon Q Developer</td><td><code>spm connect amazonq</code></td></tr><tr><td>Gemini CLI</td><td><code>spm connect gemini</code></td></tr><tr><td>OpenCode</td><td><code>spm connect opencode</code></td></tr><tr><td>OpenClaw</td><td><code>spm connect openclaw</code></td></tr></tbody></table>
<h2>Quick Start</h2>
<pre><code class="language-bash"># Initialize skills directory
spm init
# Search for skills
spm search "code review"
# Install a skill
spm add author/skill-name
# Connect to your AI client
spm connect claude
</code></pre>
<h2>What is a Skill?</h2>
<p>A skill is a portable, versioned package containing prompts, tools, or instructions that any AI model can use.</p>
<pre><code>my-skill/
├── skill.json # Manifest (name, version, triggers, dependencies)
└── SKILL.md # Main prompt / instructions
</code></pre>
<h2>Personas</h2>
<p>Define AI agent personalities with character traits, model settings, and skill dependencies:</p>
<pre><code class="language-bash"># Create a persona
spm persona create my-agent
# Activate (auto-installs missing skills)
spm persona activate my-agent
# Export to external platform
spm persona export my-agent -f openclaw
# Deploy
spm persona deploy my-agent -t openclaw
</code></pre>
<h2>Commands</h2>
<h3>Manage Skills</h3>
<table><thead><tr><th>Command</th><th>Description</th></tr></thead><tbody><tr><td><code>spm add <ref></code></td><td>Install a skill</td></tr><tr><td><code>spm install</code></td><td>Install all dependencies from skill.json</td></tr><tr><td><code>spm remove <ref></code></td><td>Remove a skill</td></tr><tr><td><code>spm create <name></code></td><td>Scaffold a new skill</td></tr><tr><td><code>spm link <path></code></td><td>Symlink a local skill for development</td></tr><tr><td><code>spm convert <file></code></td><td>Convert .md/.txt prompts into skills</td></tr><tr><td><code>spm list</code></td><td>List installed skills</td></tr><tr><td><code>spm info <name></code></td><td>Show skill details</td></tr><tr><td><code>spm validate</code></td><td>Validate a skill directory</td></tr></tbody></table>
<h3>Personas</h3>
<table><thead><tr><th>Command</th><th>Description</th></tr></thead><tbody><tr><td><code>spm persona create</code></td><td>Create a new persona</td></tr><tr><td><code>spm persona list</code></td><td>List installed personas</td></tr><tr><td><code>spm persona activate</code></td><td>Activate persona (auto-installs skills)</td></tr><tr><td><code>spm persona deactivate</code></td><td>Deactivate current persona</td></tr><tr><td><code>spm persona export</code></td><td>Export to target platform format</td></tr><tr><td><code>spm persona deploy</code></td><td>Deploy to target platform</td></tr><tr><td><code>spm persona import</code></td><td>Import from external platform</td></tr></tbody></table>
<h3>Registry</h3>
<table><thead><tr><th>Command</th><th>Description</th></tr></thead><tbody><tr><td><code>spm search <query></code></td><td>Search local and remote registries</td></tr><tr><td><code>spm publish <path></code></td><td>Publish to registry</td></tr><tr><td><code>spm update <path></code></td><td>Update a published skill</td></tr><tr><td><code>spm login</code></td><td>Authenticate (GitHub OAuth)</td></tr><tr><td><code>spm rate <name></code></td><td>Rate a skill (1-5)</td></tr><tr><td><code>spm registry add <url></code></td><td>Add a remote registry</td></tr></tbody></table>
<h3>System</h3>
<table><thead><tr><th>Command</th><th>Description</th></tr></thead><tbody><tr><td><code>spm connect <client></code></td><td>Connect MCP server to AI client</td></tr><tr><td><code>spm disconnect <client></code></td><td>Disconnect from AI client</td></tr><tr><td><code>spm serve</code></td><td>Start MCP server (stdio)</td></tr><tr><td><code>spm init</code></td><td>Initialize skills directory</td></tr><tr><td><code>spm reindex</code></td><td>Rebuild skill index</td></tr></tbody></table>
<h2>MCP Server</h2>
<p>Skillbase includes a built-in <a href="https://modelcontextprotocol.io">Model Context Protocol</a> server that exposes installed skills as tools to any MCP-compatible AI client:</p>
<pre><code class="language-bash">spm serve --stdio
</code></pre>
<p><strong>MCP tools provided:</strong> <code>skill_list</code>, <code>skill_load</code>, <code>skill_search</code>, <code>skill_context</code>, <code>skill_feedback</code>, <code>skill_install</code>, <code>persona_list</code>, <code>persona_load</code></p>
<h2>Registries</h2>
<p>Skills can be published to self-hosted registries or installed directly from GitHub:</p>
<pre><code class="language-bash"># Install from registry
spm add author/skill-name
# Install from GitHub
spm add github:author/repo
# Add a custom registry
spm registry add https://registry.example.com
# Publish
spm publish ./my-skill
</code></pre>
<h2>Requirements</h2>
<ul>
<li>Node.js >= 20.0.0</li>
</ul>
<h2>License</h2>
<p><a href="LICENSE">MIT</a></p> | https://mcpservers.org/servers/useskillbase/spm | https://mcpservers.org/all?sort=newest&page=1 |
cesium-mcp | AI-powered CesiumJS 3D globe control — 43 tools for camera, entities, layers, animation, and interaction via MCP protocol. Also available as a remote server via Streamable HTTP. | https://github.com/gaopengbin/cesium-mcp | null | <div align="center"> <img src="docs/public/logo.svg" alt="Cesium MCP" width="120"> <h1>Cesium MCP</h1> <p><strong>AI-Powered 3D Globe Control via Model Context Protocol</strong></p> <p>Connect any MCP-compatible AI agent to <a href="https://cesium.com/">CesiumJS</a> — camera, layers, entities, spatial analysis, all through natural language.</p> <p> <a href="https://gaopengbin.github.io/cesium-mcp/">Website</a> · <a href="README.zh-CN.md">中文</a> · <a href="https://gaopengbin.github.io/cesium-mcp/guide/getting-started.html">Getting Started</a> · <a href="https://gaopengbin.github.io/cesium-mcp/api/bridge.html">API Reference</a> </p> <p> <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-blue.svg" alt="License: MIT"></a> <a href="https://github.com/gaopengbin/cesium-mcp/actions/workflows/ci.yml"><img src="https://github.com/gaopengbin/cesium-mcp/actions/workflows/ci.yml/badge.svg" alt="CI"></a> <a href="https://www.npmjs.com/package/cesium-mcp-bridge"><img src="https://img.shields.io/npm/v/cesium-mcp-bridge?label=bridge" alt="npm bridge"></a> <a href="https://www.npmjs.com/package/cesium-mcp-runtime"><img src="https://img.shields.io/npm/v/cesium-mcp-runtime?label=runtime" alt="npm runtime"></a> <a href="https://www.npmjs.com/package/cesium-mcp-dev"><img src="https://img.shields.io/npm/v/cesium-mcp-dev?label=dev" alt="npm dev"></a> <a href="https://glama.ai/mcp/servers/gaopengbin/cesium-mcp"><img src="https://glama.ai/mcp/servers/gaopengbin/cesium-mcp/badges/card.svg" alt="cesium-mcp MCP server"></a> </p> </div> Demo https://github.com/user-attachments/assets/8a40565a-fcdd-47bf-ae67-bc870611c908 Packages Package Description npm cesium-mcp-bridge Browser SDK — embeds in your CesiumJS app, receives commands via WebSocket cesium-mcp-runtime MCP Server (stdio) — 49 tools (11 toolsets) + 2 resources, dynamic discovery cesium-mcp-dev IDE MCP Server — CesiumJS API helper for coding assistants Architecture ┌──────────────┐ stdio ┌──────────────────┐ WebSocket ┌──────────────────┐ │ AI Agent │ ◄────────► │ cesium-mcp- │ ◄─────────► │ cesium-mcp- │ │ (Claude, │ MCP │ runtime │ JSON-RPC │ bridge │ │ Cursor…) │ │ (Node.js) │ │ (Browser) │ └──────────────┘ └──────────────────┘ └──────────────────┘ │ ┌──────▼──────┐ │ CesiumJS │ │ Viewer │ └─────────────┘ Quick Start 1. Install the bridge in your CesiumJS app npm install cesium-mcp-bridge import { CesiumMcpBridge } from 'cesium-mcp-bridge'; const bridge = new CesiumMcpBridge(viewer, { port: 9100 }); bridge.connect(); 2. Start the MCP runtime npx cesium-mcp-runtime 3. Connect your AI agent Add to your MCP client config (e.g. Claude Desktop): { "mcpServers": { "cesium": { "command": "npx", "args": ["-y", "cesium-mcp-runtime"] } } } Now ask your AI: "Fly to the Eiffel Tower and add a red marker" 49 Available Tools Tools are organized into 11 toolsets . Default mode enables 4 core toolsets (~24 tools). Set CESIUM_TOOLSETS=all for everything, or let the AI discover and activate toolsets dynamically at runtime. Toolset Tools view (default) flyTo , setView , getView , zoomToExtent , saveViewpoint , loadViewpoint , listViewpoints entity (default) addMarker , addLabel , addModel , addPolygon , addPolyline , updateEntity , removeEntity , batchAddEntities , queryEntities layer (default) addGeoJsonLayer , listLayers , removeLayer , setLayerVisibility , updateLayerStyle , setBasemap interaction (default) screenshot , highlight camera lookAtTransform , startOrbit , stopOrbit , setCameraOptions entity-ext addBillboard , addBox , addCorridor , addCylinder , addEllipse , addRectangle , addWall animation createAnimation , controlAnimation , removeAnimation , listAnimations , updateAnimationPath , trackEntity , controlClock , setGlobeLighting tiles load3dTiles , loadTerrain , loadImageryService trajectory playTrajectory heatmap addHeatmap geolocation geocode Relationship with CesiumGS official MCP servers : The camera , entity-ext , and animation toolsets natively fuse capabilities from CesiumGS/cesium-mcp-server (Camera Server, Entity Server, Animation Server) into this project's unified bridge architecture. This means you get all official functionality plus additional tools — in a single MCP server, without running multiple processes. Examples See examples/minimal/ for a complete working demo. Development git clone https://github.com/gaopengbin/cesium-mcp.git cd cesium-mcp npm install npm run build Version Policy The major.minor version tracks CesiumJS (e.g. 1.139.x targets Cesium ~1.139.0 ). Patch versions are independent for MCP feature iterations. License MIT | <div align="center">
<img src="docs/public/logo.svg" alt="Cesium MCP" width="120"><!-- -->
<!-- --> <h1>Cesium MCP</h1><!-- -->
<!-- --> <p><strong>AI-Powered 3D Globe Control via Model Context Protocol</strong></p><!-- -->
<!-- --> <p>Connect any MCP-compatible AI agent to <a href="https://cesium.com/">CesiumJS</a> — camera, layers, entities, spatial analysis, all through natural language.</p><!-- -->
<!-- --> <p>
<a href="https://gaopengbin.github.io/cesium-mcp/">Website</a> &middot;
<a href="README.zh-CN.md">中文</a> &middot;
<a href="https://gaopengbin.github.io/cesium-mcp/guide/getting-started.html">Getting Started</a> &middot;
<a href="https://gaopengbin.github.io/cesium-mcp/api/bridge.html">API Reference</a>
</p><!-- -->
<!-- --> <p>
<a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-blue.svg" alt="License: MIT"></a>
<a href="https://github.com/gaopengbin/cesium-mcp/actions/workflows/ci.yml"><img src="https://github.com/gaopengbin/cesium-mcp/actions/workflows/ci.yml/badge.svg" alt="CI"></a>
<a href="https://www.npmjs.com/package/cesium-mcp-bridge"><img src="https://img.shields.io/npm/v/cesium-mcp-bridge?label=bridge" alt="npm bridge"></a>
<a href="https://www.npmjs.com/package/cesium-mcp-runtime"><img src="https://img.shields.io/npm/v/cesium-mcp-runtime?label=runtime" alt="npm runtime"></a>
<a href="https://www.npmjs.com/package/cesium-mcp-dev"><img src="https://img.shields.io/npm/v/cesium-mcp-dev?label=dev" alt="npm dev"></a>
<a href="https://glama.ai/mcp/servers/gaopengbin/cesium-mcp"><img src="https://glama.ai/mcp/servers/gaopengbin/cesium-mcp/badges/card.svg" alt="cesium-mcp MCP server"></a>
</p>
</div><!-- -->
<hr>
<h2>Demo</h2>
<p><a href="https://github.com/user-attachments/assets/8a40565a-fcdd-47bf-ae67-bc870611c908">https://github.com/user-attachments/assets/8a40565a-fcdd-47bf-ae67-bc870611c908</a></p>
<h2>Packages</h2>
<table><thead><tr><th>Package</th><th>Description</th><th>npm</th></tr></thead><tbody><tr><td><a href="packages/cesium-mcp-bridge/">cesium-mcp-bridge</a></td><td>Browser SDK — embeds in your CesiumJS app, receives commands via WebSocket</td><td><a href="https://www.npmjs.com/package/cesium-mcp-bridge"><img alt="npm" src="https://img.shields.io/npm/v/cesium-mcp-bridge"></a></td></tr><tr><td><a href="packages/cesium-mcp-runtime/">cesium-mcp-runtime</a></td><td>MCP Server (stdio) — 49 tools (11 toolsets) + 2 resources, dynamic discovery</td><td><a href="https://www.npmjs.com/package/cesium-mcp-runtime"><img alt="npm" src="https://img.shields.io/npm/v/cesium-mcp-runtime"></a></td></tr><tr><td><a href="packages/cesium-mcp-dev/">cesium-mcp-dev</a></td><td>IDE MCP Server — CesiumJS API helper for coding assistants</td><td><a href="https://www.npmjs.com/package/cesium-mcp-dev"><img alt="npm" src="https://img.shields.io/npm/v/cesium-mcp-dev"></a></td></tr></tbody></table>
<h2>Architecture</h2>
<pre><code>┌──────────────┐ stdio ┌──────────────────┐ WebSocket ┌──────────────────┐
│ AI Agent │ ◄────────► │ cesium-mcp- │ ◄─────────► │ cesium-mcp- │
│ (Claude, │ MCP │ runtime │ JSON-RPC │ bridge │
│ Cursor…) │ │ (Node.js) │ │ (Browser) │
└──────────────┘ └──────────────────┘ └──────────────────┘
│
┌──────▼──────┐
│ CesiumJS │
│ Viewer │
└─────────────┘
</code></pre>
<h2>Quick Start</h2>
<h3>1. Install the bridge in your CesiumJS app</h3>
<pre><code class="language-bash">npm install cesium-mcp-bridge
</code></pre>
<pre><code class="language-js">import { CesiumMcpBridge } from 'cesium-mcp-bridge';
const bridge = new CesiumMcpBridge(viewer, { port: 9100 });
bridge.connect();
</code></pre>
<h3>2. Start the MCP runtime</h3>
<pre><code class="language-bash">npx cesium-mcp-runtime
</code></pre>
<h3>3. Connect your AI agent</h3>
<p>Add to your MCP client config (e.g. Claude Desktop):</p>
<pre><code class="language-json">{
"mcpServers": {
"cesium": {
"command": "npx",
"args": ["-y", "cesium-mcp-runtime"]
}
}
}
</code></pre>
<p>Now ask your AI: <em>"Fly to the Eiffel Tower and add a red marker"</em></p>
<h2>49 Available Tools</h2>
<p>Tools are organized into <strong>11 toolsets</strong>. Default mode enables 4 core toolsets (~24 tools). Set <code>CESIUM_TOOLSETS=all</code> for everything, or let the AI discover and activate toolsets dynamically at runtime.</p>
<table><thead><tr><th>Toolset</th><th>Tools</th></tr></thead><tbody><tr><td><strong>view</strong> (default)</td><td><code>flyTo</code>, <code>setView</code>, <code>getView</code>, <code>zoomToExtent</code>, <code>saveViewpoint</code>, <code>loadViewpoint</code>, <code>listViewpoints</code></td></tr><tr><td><strong>entity</strong> (default)</td><td><code>addMarker</code>, <code>addLabel</code>, <code>addModel</code>, <code>addPolygon</code>, <code>addPolyline</code>, <code>updateEntity</code>, <code>removeEntity</code>, <code>batchAddEntities</code>, <code>queryEntities</code></td></tr><tr><td><strong>layer</strong> (default)</td><td><code>addGeoJsonLayer</code>, <code>listLayers</code>, <code>removeLayer</code>, <code>setLayerVisibility</code>, <code>updateLayerStyle</code>, <code>setBasemap</code></td></tr><tr><td><strong>interaction</strong> (default)</td><td><code>screenshot</code>, <code>highlight</code></td></tr><tr><td>camera</td><td><code>lookAtTransform</code>, <code>startOrbit</code>, <code>stopOrbit</code>, <code>setCameraOptions</code></td></tr><tr><td>entity-ext</td><td><code>addBillboard</code>, <code>addBox</code>, <code>addCorridor</code>, <code>addCylinder</code>, <code>addEllipse</code>, <code>addRectangle</code>, <code>addWall</code></td></tr><tr><td>animation</td><td><code>createAnimation</code>, <code>controlAnimation</code>, <code>removeAnimation</code>, <code>listAnimations</code>, <code>updateAnimationPath</code>, <code>trackEntity</code>, <code>controlClock</code>, <code>setGlobeLighting</code></td></tr><tr><td>tiles</td><td><code>load3dTiles</code>, <code>loadTerrain</code>, <code>loadImageryService</code></td></tr><tr><td>trajectory</td><td><code>playTrajectory</code></td></tr><tr><td>heatmap</td><td><code>addHeatmap</code></td></tr><tr><td>geolocation</td><td><code>geocode</code></td></tr></tbody></table>
<blockquote>
<p><strong>Relationship with CesiumGS official MCP servers</strong>: The <code>camera</code>, <code>entity-ext</code>, and <code>animation</code> toolsets natively fuse capabilities from <a href="https://github.com/CesiumGS/cesium-mcp-server">CesiumGS/cesium-mcp-server</a> (Camera Server, Entity Server, Animation Server) into this project's unified bridge architecture. This means you get all official functionality plus additional tools — in a single MCP server, without running multiple processes.</p>
</blockquote>
<h2>Examples</h2>
<p>See <a href="examples/minimal/">examples/minimal/</a> for a complete working demo.</p>
<h2>Development</h2>
<pre><code class="language-bash">git clone https://github.com/gaopengbin/cesium-mcp.git
cd cesium-mcp
npm install
npm run build
</code></pre>
<h2>Version Policy</h2>
<p>The major.minor version tracks CesiumJS (e.g. <code>1.139.x</code> targets Cesium <code>~1.139.0</code>). Patch versions are independent for MCP feature iterations.</p>
<h2>License</h2>
<p><a href="LICENSE">MIT</a></p> | https://mcpservers.org/servers/gaopengbin/cesium-mcp | https://mcpservers.org/all?sort=newest&page=1 |
jCodeMunch-MCP | Token-efficient MCP server for GitHub source code exploration via tree-sitter AST parsing | https://github.com/jgravelle/jcodemunch-mcp | jCodeMunch MCP | Quickstart - https://github.com/jgravelle/jcodemunch-mcp/blob/main/QUICKSTART.md FREE FOR PERSONAL USE Use it to make money, and Uncle J. gets a taste. Fair enough? details Cut code-reading token costs by up to 99% Most AI agents explore repositories the expensive way: open entire files → skim thousands of irrelevant lines → repeat. jCodeMunch indexes a codebase once and lets agents retrieve only the exact symbols they need — functions, classes, methods, constants — with byte-level precision. Task Traditional approach With jCodeMunch Find a function ~40,000 tokens ~200 tokens Understand module API ~15,000 tokens ~800 tokens Explore repo structure ~200,000 tokens ~2k tokens Index once. Query cheaply forever. Precision context beats brute-force context. jCodeMunch MCP Structured retrieval for serious AI agents Commercial licenses jCodeMunch-MCP is free for non-commercial use . Commercial use requires a paid license. jCodeMunch-only licenses Builder — $79 — 1 developer Studio — $349 — up to 5 developers Platform — $1,999 — org-wide internal deployment Want both code and docs retrieval? Munch Duo Builder Bundle — $89 Munch Duo Studio Bundle — $399 Munch Duo Platform Bundle — $2,249 Stop dumping files into context windows. Start retrieving exactly what the agent needs. jCodeMunch indexes a codebase once using tree-sitter AST parsing, then allows MCP-compatible agents (Claude Desktop, VS Code, Google Antigravity, and others) to discover and retrieve code by symbol instead of brute-reading files. Every symbol stores: Signature Kind Qualified name One-line summary Byte offsets into the original file Full source is retrieved on demand using O(1) byte-offset seeking. Proof: Token savings in the wild Repo: geekcomputers/Python Size: 338 files, 1,422 symbols indexed Task: Locate calculator / math implementations Approach Tokens What the agent had to do Raw file approach ~7,500 Open multiple files and scan manually jCodeMunch MCP ~1,449 search_symbols() → get_symbol() Result: ~80% fewer tokens (~5× more efficient) Cost scales with tokens. Latency scales with irrelevant context. jCodeMunch turns search into navigation. Why agents need this Agents waste money when they: Open entire files to find one function Re-read the same code repeatedly Consume imports, boilerplate, and unrelated helpers jCodeMunch provides precision context access: Search symbols by name, kind, or language Outline files without loading full contents Retrieve exact symbol implementations only Fall back to full-text search when necessary Agents do not need larger context windows. They need structured retrieval. How it works jCodeMunch implements jMRI-Full — the open specification for structured retrieval MCP servers. jMRI-Full covers the full stack: discover, search, retrieve, and metadata operations with batch retrieval, hash-based drift detection, byte-offset addressing, and a complete _meta envelope on every call. Discovery — GitHub API or local directory walk Security filtering — traversal protection, secret exclusion, binary detection Parsing — tree-sitter AST extraction Context enrichment — auto-detected ecosystem providers (dbt, etc.) inject business metadata Storage — JSON index + raw files stored locally ( ~/.code-index/ ) Retrieval — O(1) byte-offset seeking via stable symbol IDs Stable Symbol IDs {file_path}::{qualified_name}#{kind} Examples: src/main.py::UserService.login#method src/utils.py::authenticate#function IDs remain stable across re-indexing when path, qualified name, and kind are unchanged. Installation New here? See QUICKSTART.md for a focused 3-step setup guide. Prerequisites Python 3.10+ pip Install pip install jcodemunch-mcp Verify: jcodemunch-mcp --help Configure MCP Client PATH note: MCP clients often run with a limited environment where jcodemunch-mcp may not be found even if it works in your terminal. Using uvx is the recommended approach — it resolves the package on demand without requiring anything to be on your system PATH. If you prefer pip install , use the absolute path to the executable instead: Linux: /home/<username>/.local/bin/jcodemunch-mcp macOS: /Users/<username>/.local/bin/jcodemunch-mcp Windows: C:\\Users\\<username>\\AppData\\Roaming\\Python\\Python3xx\\Scripts\\jcodemunch-mcp.exe Claude Code The fastest way to add jCodeMunch to Claude Code is a single command: claude mcp add jcodemunch uvx jcodemunch-mcp This registers the server at user scope ( ~/.claude.json ) so it is available in every project. To add it to a specific project only, pass --scope project : claude mcp add --scope project jcodemunch uvx jcodemunch-mcp To include optional environment variables (e.g. GITHUB_TOKEN or ANTHROPIC_API_KEY ): claude mcp add jcodemunch uvx jcodemunch-mcp \ -e GITHUB_TOKEN=ghp_... \ -e ANTHROPIC_API_KEY=sk-ant-... Restart Claude Code after adding the server. Manual config — if you prefer to edit the config file directly, the relevant files are: Scope Path User (global) ~/.claude.json Project .claude/settings.json (in the project root) { "mcpServers": { "jcodemunch": { "command": "uvx", "args": ["jcodemunch-mcp"] } } } Claude Desktop Config file location: OS Path macOS ~/Library/Application Support/Claude/claude_desktop_config.json Linux ~/.config/claude/claude_desktop_config.json Windows %APPDATA%\Claude\claude_desktop_config.json Minimal config (no API keys needed): { "mcpServers": { "jcodemunch": { "command": "uvx", "args": ["jcodemunch-mcp"] } } } With optional AI summaries and GitHub auth: { "mcpServers": { "jcodemunch": { "command": "uvx", "args": ["jcodemunch-mcp"], "env": { "GITHUB_TOKEN": "ghp_...", "ANTHROPIC_API_KEY": "sk-ant-..." } } } } With debug logging (useful when diagnosing why files are not indexed): { "mcpServers": { "jcodemunch": { "command": "uvx", "args": [ "jcodemunch-mcp", "--log-level", "DEBUG", "--log-file", "/tmp/jcodemunch.log" ] } } } Logging flags can also be set via env vars JCODEMUNCH_LOG_LEVEL and JCODEMUNCH_LOG_FILE . Always use --log-file (or the env var) when debugging — writing logs to stderr can corrupt the MCP stdio stream in some clients. After saving the config, restart Claude Desktop for the server to appear. Google Antigravity Open the Agent pane → click the ⋯ menu → MCP Servers → Manage MCP Servers Click View raw config to open mcp_config.json Add the entry below, save, then restart the MCP server from the Manage MCPs pane { "mcpServers": { "jcodemunch": { "command": "uvx", "args": ["jcodemunch-mcp"] } } } Environment variables are optional: Variable Purpose GITHUB_TOKEN Higher GitHub API limits / private access ANTHROPIC_API_KEY AI-generated summaries via Claude Haiku (takes priority) ANTHROPIC_BASE_URL Third-party Anthropic-compatible endpoints (e.g. z.ai) GOOGLE_API_KEY AI-generated summaries via Gemini Flash Step 3: Tell Claude to actually use it This step is not optional. Installing the MCP server makes the tools available — but Claude will not use them automatically. Without instructions, Claude defaults to its built-in file tools (read, grep, etc.) and never touches jCodeMunch. This is the single most common reason users install the server and see no difference. Create a CLAUDE.md file that instructs Claude to use jCodeMunch for all code lookups. Global (applies to every project) Create or edit ~/.claude/CLAUDE.md : Use jcodemunch-mcp for all code lookups. Never read full files when MCP is available. 1. Call `list_repos` first — if the project is not indexed, call `index_folder` with the current directory. 2. Use `search_symbols` / `get_symbol` to find and retrieve code by symbol name. 3. Use `get_repo_outline` or `get_file_outline` to explore structure. 4. Fall back to direct file reads only when editing or when MCP is unavailable. Project-level only Create CLAUDE.md in your project root with the same content. Claude Code merges project-level and global instructions automatically. Verify it's working Ask Claude: "What repos do you have indexed?" — it should call list_repos . If it responds without calling any tool, re-check that CLAUDE.md exists and that the MCP server appears in /mcp (Claude Code) or the server list in Claude Desktop. Usage Examples index_folder: { "path": "/path/to/project" } index_repo: { "url": "owner/repo" } get_repo_outline: { "repo": "owner/repo" } get_file_outline: { "repo": "owner/repo", "file_path": "src/main.py" } get_file_content: { "repo": "owner/repo", "file_path": "src/main.py", "start_line": 10, "end_line": 25 } search_symbols: { "repo": "owner/repo", "query": "authenticate" } get_symbol: { "repo": "owner/repo", "symbol_id": "src/main.py::MyClass.login#method" } get_context_bundle: { "repo": "owner/repo", "symbol_id": "src/main.py::MyClass.login#method" } search_text: { "repo": "owner/repo", "query": "TODO", "context_lines": 1 } search_columns: { "repo": "owner/repo", "query": "customer_id", "model_pattern": "fact_*" } Local folder indexes are stored with stable hashed repo ids. Use list_repos to inspect the exact id, or the bare display name when it is unique. Tools (14) Tool Purpose index_repo Index a GitHub repository index_folder Index a local folder list_repos List indexed repositories get_file_tree Repository file structure get_file_outline Symbol hierarchy for a file get_file_content Retrieve cached file content get_symbol Retrieve full symbol source get_symbols Batch retrieve symbols get_context_bundle Symbol source + file imports in one call search_symbols Search symbols with filters search_text Full-text search with context search_columns Search column metadata across models get_repo_outline High-level repo overview invalidate_cache Remove cached index Every tool response includes a _meta envelope with timing, token savings, and cost avoided: "_meta": { "timing_ms": 4.3, "tokens_saved": 48153, "total_tokens_saved": 1280837, "cost_avoided": { "claude_opus": 1.2038, "gpt5_latest": 0.4815 }, "total_cost_avoided": { "claude_opus": 32.02, "gpt5_latest": 12.81 } } total_tokens_saved and total_cost_avoided accumulate across all tool calls and persist to ~/.code-index/_savings.json . Recent Updates v1.4.1 — CLI interface ( cli/cli.py ) for terminal/pipeline use; "Tell Claude to use it" setup section in README v0.2.10 — Pin mcp<1.10.0 to prevent Windows win32api DLL crash on startup v0.2.9 — Community savings meter: anonymous token savings shared to a live global counter at j.gravelle.us (opt-out via JCODEMUNCH_SHARE_SAVINGS=0 ); updated model pricing (Opus $25/1M, GPT-5 $10/1M) v0.2.8 — Estimated cost avoided added to every _meta response ( cost_avoided , total_cost_avoided ) v0.2.7 — Security fix: .claude/ excluded from sdist; structural CI guardrails prevent credential bundling v0.2.5 — Path traversal hardening in IndexStore ; jcodemunch-mcp --help now works v0.2.4 — Live token savings counter ( tokens_saved , total_tokens_saved in every _meta ) v0.2.3 — Google Gemini Flash support ( GOOGLE_API_KEY ); auto-selects between Anthropic and Gemini v0.2.2 — PHP language support Supported Languages Language Extensions Symbol Types Python .py function, class, method, constant, type JavaScript .js , .jsx function, class, method, constant TypeScript .ts , .tsx function, class, method, constant, type Go .go function, method, type, constant Rust .rs function, type, impl, constant Java .java method, class, type, constant PHP .php function, class, method, type, constant Dart .dart function, class, method, type C# .cs class, method, type, record C .c function, type, constant C++ .cpp , .cc , .cxx , .hpp , .hh , .hxx , .h * function, class, method, type, constant Elixir .ex , .exs class (module/impl), type (protocol/@type/@callback), method, function Ruby .rb , .rake class, type (module), method, function SQL .sql function (CREATE FUNCTION, CTE, dbt macro/test/materialization), type (CREATE TABLE/VIEW/SCHEMA/INDEX, dbt snapshot) XML/XUL .xml , .xul type (root element), constant (id attributes), function (script refs) * .h is parsed as C++ first, then falls back to C when no C++ symbols are extracted. See LANGUAGE_SUPPORT.md for full semantics. Context Providers When indexing local folders, jCodeMunch automatically detects ecosystem tools and enriches the index with business context — descriptions, tags, and metadata from project configuration files. Provider Detects Enriches With dbt dbt_project.yml Model descriptions, tags, column names/descriptions Context enrichment is automatic — no configuration needed. When a provider detects its tool, it injects metadata into AI summarization prompts, file summaries, and search keywords. Example: a dbt model with a schema.yml description produces file summaries like: This table summarizes account ledger. Tags: nightly, agg, intraday. 70 properties Instead of the default: Contains 2 functions: source, renamed The provider system is extensible — adding support for Terraform, OpenAPI, Django, or any other tool requires implementing a single ContextProvider class. See CONTEXT_PROVIDERS.md for the full architecture, dbt details, and guide to writing new providers. Contributing PRs welcome! All contributors must sign the Contributor License Agreement before their PR can be merged — CLA Assistant will prompt you automatically. See CONTRIBUTING.md for details. Security Built-in protections: Path traversal prevention (owner/name sanitization + _safe_content_path enforcement) Symlink escape protection Secret file exclusion ( .env , *.pem , etc.) Binary detection Configurable file size limits See SECURITY.md for details. Best Use Cases Large multi-module repositories Agent-driven refactors Architecture exploration Faster onboarding Token-efficient multi-agent workflows Not Intended For LSP diagnostics or completions Editing workflows Real-time file watching Cross-repository global indexing Semantic program analysis Local LLMs (Ollama / LM Studio) You can use local, privacy-preserving AI models to generate summaries by providing an OpenAI-compatible endpoint. For Ollama , run a model locally, then configure the MCP server: "env": { "OPENAI_API_BASE": "http://localhost:11434/v1", "OPENAI_MODEL": "qwen3-coder" } For LM Studio , ensure the Local Server is running (usually on port 1234): "env": { "OPENAI_API_BASE": "http://127.0.0.1:1234/v1", "OPENAI_MODEL": "openai/gpt-oss-20b" } [!TIP] Performance Note: Local models can be slow to load into memory on their first request, potentially causing the MCP server to time out and fall back to generic signature summaries. It is highly recommended to pre-load the model in Ollama or LM Studio before starting the server, or increase the OPENAI_TIMEOUT environment variable (e.g., to "120.0" ) to allow more time for generation. Environment Variables Variable Purpose Required GITHUB_TOKEN GitHub API auth No ANTHROPIC_API_KEY Symbol summaries via Claude Haiku (takes priority) No ANTHROPIC_BASE_URL Third-party Anthropic-compatible endpoints (e.g. z.ai) No ANTHROPIC_MODEL Model name for Claude summaries (default: claude-haiku-4-5-20251001 ) No GOOGLE_API_KEY Symbol summaries via Gemini Flash No GOOGLE_MODEL Model name for Gemini summaries (default: gemini-2.5-flash-lite ) No OPENAI_API_BASE Base URL for local LLMs (e.g. http://localhost:11434/v1 ) No OPENAI_API_KEY API key for local LLMs (default: local-llm ) No OPENAI_MODEL Model name for local LLMs (default: qwen3-coder ) No OPENAI_TIMEOUT Timeout in seconds for local requests (default: 60.0 ) No OPENAI_BATCH_SIZE Symbols per summarization request (default: 10 ) No OPENAI_CONCURRENCY Max parallel batch requests (default: 1 ) No OPENAI_MAX_TOKENS Max output tokens per batch response (default: 500 ) No CODE_INDEX_PATH Custom cache path No JCODEMUNCH_MAX_INDEX_FILES Maximum files to index per repo/folder (default: 10000 ) No JCODEMUNCH_CONTEXT_PROVIDERS Set to 0 to disable context providers (dbt, etc.) during indexing No JCODEMUNCH_SHARE_SAVINGS Set to 0 to disable anonymous community token savings reporting No JCODEMUNCH_LOG_LEVEL Log level: DEBUG , INFO , WARNING , ERROR (default: WARNING ) No JCODEMUNCH_LOG_FILE Path to log file. If unset, logs go to stderr. Use a file to avoid polluting MCP stdio. No Community Savings Meter Each tool call contributes an anonymous delta to a live global counter at j.gravelle.us . Only two values are ever sent: the tokens saved (a number) and a random anonymous install ID — never code, paths, repo names, or anything identifying. The anon ID is generated once and stored in ~/.code-index/_savings.json . To disable, set JCODEMUNCH_SHARE_SAVINGS=0 in your MCP server env. Documentation USER_GUIDE.md ARCHITECTURE.md SPEC.md SECURITY.md LANGUAGE_SUPPORT.md CONTEXT_PROVIDERS.md Star History <a href="https://www.star-history.com/#jgravelle/jcodemunch-mcp&type=date&legend=top-left"> <picture> <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=jgravelle/jcodemunch-mcp&type=date&theme=dark&legend=top-left" /> <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/svg?repos=jgravelle/jcodemunch-mcp&type=date&legend=top-left" /> <img alt="Star History Chart" src="https://api.star-history.com/svg?repos=jgravelle/jcodemunch-mcp&type=date&legend=top-left" /> </picture> </a> License (Dual Use) This repository is free for non-commercial use under the terms below. Commercial use requires a paid commercial license. Copyright and License Text Copyright (c) 2026 J. Gravelle 1. Non-Commercial License Grant (Free) Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to use, copy, modify, merge, publish, and distribute the Software for personal, educational, research, hobby, or other non-commercial purposes , subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. Any modifications made to the Software must clearly indicate that they are derived from the original work, and the name of the original author (J. Gravelle) must remain intact. He's kinda full of himself. Redistributions of the Software in source code form must include a prominent notice describing any modifications from the original version. 2. Commercial Use Commercial use of the Software requires a separate paid commercial license from the author. “Commercial use” includes, but is not limited to: Use of the Software in a business environment Internal use within a for-profit organization Incorporation into a product or service offered for sale Use in connection with revenue generation, consulting, SaaS, hosting, or fee-based services For commercial licensing inquiries, contact: [email protected] | https://j.gravelle.us Until a commercial license is obtained, commercial use is not permitted. 3. Disclaimer of Warranty THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHOR OR COPYRIGHT HOLDER BE LIABLE FOR ANY CLAIM, DAMAGES, OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT, OR OTHERWISE, ARISING FROM, OUT OF, OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. | <p>Quickstart - <a href="https://github.com/jgravelle/jcodemunch-mcp/blob/main/QUICKSTART.md">https://github.com/jgravelle/jcodemunch-mcp/blob/main/QUICKSTART.md</a></p>
<h2>FREE FOR PERSONAL USE</h2>
<p><strong>Use it to make money, and Uncle J. gets a taste. Fair enough?</strong> <a href="#commercial-licenses">details</a></p>
<h2>Cut code-reading token costs by up to <strong>99%</strong></h2>
<p>Most AI agents explore repositories the expensive way:
open entire files → skim thousands of irrelevant lines → repeat.</p>
<p><strong>jCodeMunch indexes a codebase once and lets agents retrieve only the exact symbols they need</strong> — functions, classes, methods, constants — with byte-level precision.</p>
<table><thead><tr><th>Task</th><th>Traditional approach</th><th>With jCodeMunch</th></tr></thead><tbody><tr><td>Find a function</td><td>~40,000 tokens</td><td>~200 tokens</td></tr><tr><td>Understand module API</td><td>~15,000 tokens</td><td>~800 tokens</td></tr><tr><td>Explore repo structure</td><td>~200,000 tokens</td><td>~2k tokens</td></tr></tbody></table>
<p>Index once. Query cheaply forever.<br>
<!-- -->Precision context beats brute-force context.</p>
<hr>
<h1>jCodeMunch MCP</h1>
<h3>Structured retrieval for serious AI agents</h3>
<p><img alt="License" src="https://img.shields.io/badge/license-dual--use-blue">
<img alt="MCP" src="https://img.shields.io/badge/MCP-compatible-purple">
<img alt="Local-first" src="https://img.shields.io/badge/local--first-yes-brightgreen">
<img alt="Polyglot" src="https://img.shields.io/badge/parsing-tree--sitter-9cf">
<img alt="jMRI" src="https://img.shields.io/badge/jMRI-Full-blueviolet">
<a href="https://pypi.org/project/jcodemunch-mcp/"><img alt="PyPI version" src="https://img.shields.io/pypi/v/jcodemunch-mcp"></a>
<a href="https://pypi.org/project/jcodemunch-mcp/"><img alt="PyPI - Python Version" src="https://img.shields.io/pypi/pyversions/jcodemunch-mcp"></a></p>
<blockquote>
<h2>Commercial licenses</h2>
<p>jCodeMunch-MCP is <strong>free for non-commercial use</strong>.</p>
<p><strong>Commercial use requires a paid license.</strong></p>
<p><strong>jCodeMunch-only licenses</strong></p>
<ul>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#builder">Builder — $79</a> — 1 developer</li>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#studio">Studio — $349</a> — up to 5 developers</li>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#platform">Platform — $1,999</a> — org-wide internal deployment</li>
</ul>
<p><strong>Want both code and docs retrieval?</strong></p>
<ul>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#builder">Munch Duo Builder Bundle — $89</a></li>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#studio">Munch Duo Studio Bundle — $399</a></li>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#platform">Munch Duo Platform Bundle — $2,249</a></li>
</ul>
</blockquote>
<p><strong>Stop dumping files into context windows. Start retrieving exactly what the agent needs.</strong></p>
<p>jCodeMunch indexes a codebase once using tree-sitter AST parsing, then allows MCP-compatible agents (Claude Desktop, VS Code, Google Antigravity, and others) to <strong>discover and retrieve code by symbol</strong> instead of brute-reading files.</p>
<p>Every symbol stores:</p>
<ul>
<li>Signature</li>
<li>Kind</li>
<li>Qualified name</li>
<li>One-line summary</li>
<li>Byte offsets into the original file</li>
</ul>
<p>Full source is retrieved on demand using O(1) byte-offset seeking.</p>
<hr>
<h2>Proof: Token savings in the wild</h2>
<p><strong>Repo:</strong> <code>geekcomputers/Python</code><br>
<strong>Size:</strong> 338 files, 1,422 symbols indexed<br>
<strong>Task:</strong> Locate calculator / math implementations</p>
<table><thead><tr><th>Approach</th><th style="text-align:right">Tokens</th><th>What the agent had to do</th></tr></thead><tbody><tr><td>Raw file approach</td><td style="text-align:right">~7,500</td><td>Open multiple files and scan manually</td></tr><tr><td>jCodeMunch MCP</td><td style="text-align:right">~1,449</td><td><code>search_symbols()</code> → <code>get_symbol()</code></td></tr></tbody></table>
<h3>Result: <strong>~80% fewer tokens</strong> (~5× more efficient)</h3>
<p>Cost scales with tokens.<br>
<!-- -->Latency scales with irrelevant context.</p>
<p>jCodeMunch turns search into navigation.</p>
<hr>
<h2>Why agents need this</h2>
<p>Agents waste money when they:</p>
<ul>
<li>Open entire files to find one function</li>
<li>Re-read the same code repeatedly</li>
<li>Consume imports, boilerplate, and unrelated helpers</li>
</ul>
<p>jCodeMunch provides precision context access:</p>
<ul>
<li>Search symbols by name, kind, or language</li>
<li>Outline files without loading full contents</li>
<li>Retrieve exact symbol implementations only</li>
<li>Fall back to full-text search when necessary</li>
</ul>
<p>Agents do not need larger context windows.<br>
<!-- -->They need structured retrieval.</p>
<hr>
<h2>How it works</h2>
<p>jCodeMunch implements <strong><a href="https://dev.to/jgravelle/your-ai-agent-is-dumpster-diving-through-your-code-326f">jMRI-Full</a></strong> — the open specification for structured retrieval MCP servers. jMRI-Full covers the full stack: discover, search, retrieve, and metadata operations with batch retrieval, hash-based drift detection, byte-offset addressing, and a complete <code>_meta</code> envelope on every call.</p>
<ol>
<li><strong>Discovery</strong> — GitHub API or local directory walk</li>
<li><strong>Security filtering</strong> — traversal protection, secret exclusion, binary detection</li>
<li><strong>Parsing</strong> — tree-sitter AST extraction</li>
<li><strong>Context enrichment</strong> — auto-detected ecosystem providers (dbt, etc.) inject business metadata</li>
<li><strong>Storage</strong> — JSON index + raw files stored locally (<code>~/.code-index/</code>)</li>
<li><strong>Retrieval</strong> — O(1) byte-offset seeking via stable symbol IDs</li>
</ol>
<h3>Stable Symbol IDs</h3>
<pre><code>{file_path}::{qualified_name}#{kind}
</code></pre>
<p>Examples:</p>
<ul>
<li><code>src/main.py::UserService.login#method</code></li>
<li><code>src/utils.py::authenticate#function</code></li>
</ul>
<p>IDs remain stable across re-indexing when path, qualified name, and kind are unchanged.</p>
<hr>
<h2>Installation</h2>
<blockquote>
<p><strong>New here?</strong> See <a href="QUICKSTART.md">QUICKSTART.md</a> for a focused 3-step setup guide.</p>
</blockquote>
<h3>Prerequisites</h3>
<ul>
<li>Python 3.10+</li>
<li>pip</li>
</ul>
<h3>Install</h3>
<pre><code class="language-bash">pip install jcodemunch-mcp
</code></pre>
<p>Verify:</p>
<pre><code class="language-bash">jcodemunch-mcp --help
</code></pre>
<hr>
<h2>Configure MCP Client</h2>
<blockquote>
<p><strong>PATH note:</strong> MCP clients often run with a limited environment where <code>jcodemunch-mcp</code> may not be found even if it works in your terminal. Using <a href="https://github.com/astral-sh/uv"><code>uvx</code></a> is the recommended approach — it resolves the package on demand without requiring anything to be on your system PATH. If you prefer <code>pip install</code>, use the absolute path to the executable instead:</p>
<ul>
<li><strong>Linux:</strong> <code>/home/<username>/.local/bin/jcodemunch-mcp</code></li>
<li><strong>macOS:</strong> <code>/Users/<username>/.local/bin/jcodemunch-mcp</code></li>
<li><strong>Windows:</strong> <code>C:\\Users\\<username>\\AppData\\Roaming\\Python\\Python3xx\\Scripts\\jcodemunch-mcp.exe</code></li>
</ul>
</blockquote>
<h3>Claude Code</h3>
<p>The fastest way to add jCodeMunch to Claude Code is a single command:</p>
<pre><code class="language-bash">claude mcp add jcodemunch uvx jcodemunch-mcp
</code></pre>
<p>This registers the server at user scope (<code>~/.claude.json</code>) so it is available in every project. To add it to a specific project only, pass <code>--scope project</code>:</p>
<pre><code class="language-bash">claude mcp add --scope project jcodemunch uvx jcodemunch-mcp
</code></pre>
<p>To include optional environment variables (e.g. <code>GITHUB_TOKEN</code> or <code>ANTHROPIC_API_KEY</code>):</p>
<pre><code class="language-bash">claude mcp add jcodemunch uvx jcodemunch-mcp \
-e GITHUB_TOKEN=ghp_... \
-e ANTHROPIC_API_KEY=sk-ant-...
</code></pre>
<p>Restart Claude Code after adding the server.</p>
<p><strong>Manual config</strong> — if you prefer to edit the config file directly, the relevant files are:</p>
<table><thead><tr><th>Scope</th><th>Path</th></tr></thead><tbody><tr><td>User (global)</td><td><code>~/.claude.json</code></td></tr><tr><td>Project</td><td><code>.claude/settings.json</code> (in the project root)</td></tr></tbody></table>
<pre><code class="language-json">{
"mcpServers": {
"jcodemunch": {
"command": "uvx",
"args": ["jcodemunch-mcp"]
}
}
}
</code></pre>
<h3>Claude Desktop</h3>
<p>Config file location:</p>
<table><thead><tr><th>OS</th><th>Path</th></tr></thead><tbody><tr><td>macOS</td><td><code>~/Library/Application Support/Claude/claude_desktop_config.json</code></td></tr><tr><td>Linux</td><td><code>~/.config/claude/claude_desktop_config.json</code></td></tr><tr><td>Windows</td><td><code>%APPDATA%\Claude\claude_desktop_config.json</code></td></tr></tbody></table>
<p><strong>Minimal config (no API keys needed):</strong></p>
<pre><code class="language-json">{
"mcpServers": {
"jcodemunch": {
"command": "uvx",
"args": ["jcodemunch-mcp"]
}
}
}
</code></pre>
<p><strong>With optional AI summaries and GitHub auth:</strong></p>
<pre><code class="language-json">{
"mcpServers": {
"jcodemunch": {
"command": "uvx",
"args": ["jcodemunch-mcp"],
"env": {
"GITHUB_TOKEN": "ghp_...",
"ANTHROPIC_API_KEY": "sk-ant-..."
}
}
}
}
</code></pre>
<p><strong>With debug logging (useful when diagnosing why files are not indexed):</strong></p>
<pre><code class="language-json">{
"mcpServers": {
"jcodemunch": {
"command": "uvx",
"args": [
"jcodemunch-mcp",
"--log-level", "DEBUG",
"--log-file", "/tmp/jcodemunch.log"
]
}
}
}
</code></pre>
<blockquote>
<p>Logging flags can also be set via env vars <code>JCODEMUNCH_LOG_LEVEL</code> and <code>JCODEMUNCH_LOG_FILE</code>. Always use <code>--log-file</code> (or the env var) when debugging — writing logs to stderr can corrupt the MCP stdio stream in some clients.</p>
</blockquote>
<p>After saving the config, <strong>restart Claude Desktop</strong> for the server to appear.</p>
<h3>Google Antigravity</h3>
<ol>
<li>Open the Agent pane → click the <code>⋯</code> menu → <strong>MCP Servers</strong> → <strong>Manage MCP Servers</strong></li>
<li>Click <strong>View raw config</strong> to open <code>mcp_config.json</code></li>
<li>Add the entry below, save, then restart the MCP server from the Manage MCPs pane</li>
</ol>
<pre><code class="language-json">{
"mcpServers": {
"jcodemunch": {
"command": "uvx",
"args": ["jcodemunch-mcp"]
}
}
}
</code></pre>
<p>Environment variables are optional:</p>
<table><thead><tr><th>Variable</th><th>Purpose</th></tr></thead><tbody><tr><td><code>GITHUB_TOKEN</code></td><td>Higher GitHub API limits / private access</td></tr><tr><td><code>ANTHROPIC_API_KEY</code></td><td>AI-generated summaries via Claude Haiku (takes priority)</td></tr><tr><td><code>ANTHROPIC_BASE_URL</code></td><td>Third-party Anthropic-compatible endpoints (e.g. z.ai)</td></tr><tr><td><code>GOOGLE_API_KEY</code></td><td>AI-generated summaries via Gemini Flash</td></tr></tbody></table>
<hr>
<h2>Step 3: Tell Claude to actually use it</h2>
<blockquote>
<p><strong>This step is not optional.</strong></p>
<p>Installing the MCP server makes the tools <em>available</em> — but Claude will not use them automatically. Without instructions, Claude defaults to its built-in file tools (read, grep, etc.) and never touches jCodeMunch. This is the single most common reason users install the server and see no difference.</p>
</blockquote>
<p>Create a <code>CLAUDE.md</code> file that instructs Claude to use jCodeMunch for all code lookups.</p>
<h3>Global (applies to every project)</h3>
<p>Create or edit <code>~/.claude/CLAUDE.md</code>:</p>
<pre><code class="language-markdown">Use jcodemunch-mcp for all code lookups. Never read full files when MCP is available.
1. Call `list_repos` first — if the project is not indexed, call `index_folder` with the current directory.
2. Use `search_symbols` / `get_symbol` to find and retrieve code by symbol name.
3. Use `get_repo_outline` or `get_file_outline` to explore structure.
4. Fall back to direct file reads only when editing or when MCP is unavailable.
</code></pre>
<h3>Project-level only</h3>
<p>Create <code>CLAUDE.md</code> in your project root with the same content. Claude Code merges project-level and global instructions automatically.</p>
<h3>Verify it's working</h3>
<p>Ask Claude: <em>"What repos do you have indexed?"</em> — it should call <code>list_repos</code>. If it responds without calling any tool, re-check that <code>CLAUDE.md</code> exists and that the MCP server appears in <code>/mcp</code> (Claude Code) or the server list in Claude Desktop.</p>
<hr>
<h2>Usage Examples</h2>
<pre><code>index_folder: { "path": "/path/to/project" }
index_repo: { "url": "owner/repo" }
get_repo_outline: { "repo": "owner/repo" }
get_file_outline: { "repo": "owner/repo", "file_path": "src/main.py" }
get_file_content: { "repo": "owner/repo", "file_path": "src/main.py", "start_line": 10, "end_line": 25 }
search_symbols: { "repo": "owner/repo", "query": "authenticate" }
get_symbol: { "repo": "owner/repo", "symbol_id": "src/main.py::MyClass.login#method" }
get_context_bundle: { "repo": "owner/repo", "symbol_id": "src/main.py::MyClass.login#method" }
search_text: { "repo": "owner/repo", "query": "TODO", "context_lines": 1 }
search_columns: { "repo": "owner/repo", "query": "customer_id", "model_pattern": "fact_*" }
</code></pre>
<p>Local folder indexes are stored with stable hashed repo ids. Use <code>list_repos</code> to inspect the exact id, or the bare display name when it is unique.</p>
<hr>
<h2>Tools (14)</h2>
<table><thead><tr><th>Tool</th><th>Purpose</th></tr></thead><tbody><tr><td><code>index_repo</code></td><td>Index a GitHub repository</td></tr><tr><td><code>index_folder</code></td><td>Index a local folder</td></tr><tr><td><code>list_repos</code></td><td>List indexed repositories</td></tr><tr><td><code>get_file_tree</code></td><td>Repository file structure</td></tr><tr><td><code>get_file_outline</code></td><td>Symbol hierarchy for a file</td></tr><tr><td><code>get_file_content</code></td><td>Retrieve cached file content</td></tr><tr><td><code>get_symbol</code></td><td>Retrieve full symbol source</td></tr><tr><td><code>get_symbols</code></td><td>Batch retrieve symbols</td></tr><tr><td><code>get_context_bundle</code></td><td>Symbol source + file imports in one call</td></tr><tr><td><code>search_symbols</code></td><td>Search symbols with filters</td></tr><tr><td><code>search_text</code></td><td>Full-text search with context</td></tr><tr><td><code>search_columns</code></td><td>Search column metadata across models</td></tr><tr><td><code>get_repo_outline</code></td><td>High-level repo overview</td></tr><tr><td><code>invalidate_cache</code></td><td>Remove cached index</td></tr></tbody></table>
<p>Every tool response includes a <code>_meta</code> envelope with timing, token savings, and cost avoided:</p>
<pre><code class="language-json">"_meta": {
"timing_ms": 4.3,
"tokens_saved": 48153,
"total_tokens_saved": 1280837,
"cost_avoided": { "claude_opus": 1.2038, "gpt5_latest": 0.4815 },
"total_cost_avoided": { "claude_opus": 32.02, "gpt5_latest": 12.81 }
}
</code></pre>
<p><code>total_tokens_saved</code> and <code>total_cost_avoided</code> accumulate across all tool calls and persist to <code>~/.code-index/_savings.json</code>.</p>
<hr>
<h2>Recent Updates</h2>
<p><strong>v1.4.1</strong> — CLI interface (<code>cli/cli.py</code>) for terminal/pipeline use; "Tell Claude to use it" setup section in README
<strong>v0.2.10</strong> — Pin <code>mcp<1.10.0</code> to prevent Windows <code>win32api</code> DLL crash on startup
<strong>v0.2.9</strong> — Community savings meter: anonymous token savings shared to a live global counter at j.gravelle.us (opt-out via <code>JCODEMUNCH_SHARE_SAVINGS=0</code>); updated model pricing (Opus $25/1M, GPT-5 $10/1M)
<strong>v0.2.8</strong> — Estimated cost avoided added to every <code>_meta</code> response (<code>cost_avoided</code>, <code>total_cost_avoided</code>)
<strong>v0.2.7</strong> — Security fix: <code>.claude/</code> excluded from sdist; structural CI guardrails prevent credential bundling
<strong>v0.2.5</strong> — Path traversal hardening in <code>IndexStore</code>; <code>jcodemunch-mcp --help</code> now works
<strong>v0.2.4</strong> — Live token savings counter (<code>tokens_saved</code>, <code>total_tokens_saved</code> in every <code>_meta</code>)
<strong>v0.2.3</strong> — Google Gemini Flash support (<code>GOOGLE_API_KEY</code>); auto-selects between Anthropic and Gemini
<strong>v0.2.2</strong> — PHP language support</p>
<hr>
<h2>Supported Languages</h2>
<table><thead><tr><th>Language</th><th>Extensions</th><th>Symbol Types</th></tr></thead><tbody><tr><td>Python</td><td><code>.py</code></td><td>function, class, method, constant, type</td></tr><tr><td>JavaScript</td><td><code>.js</code>, <code>.jsx</code></td><td>function, class, method, constant</td></tr><tr><td>TypeScript</td><td><code>.ts</code>, <code>.tsx</code></td><td>function, class, method, constant, type</td></tr><tr><td>Go</td><td><code>.go</code></td><td>function, method, type, constant</td></tr><tr><td>Rust</td><td><code>.rs</code></td><td>function, type, impl, constant</td></tr><tr><td>Java</td><td><code>.java</code></td><td>method, class, type, constant</td></tr><tr><td>PHP</td><td><code>.php</code></td><td>function, class, method, type, constant</td></tr><tr><td>Dart</td><td><code>.dart</code></td><td>function, class, method, type</td></tr><tr><td>C#</td><td><code>.cs</code></td><td>class, method, type, record</td></tr><tr><td>C</td><td><code>.c</code></td><td>function, type, constant</td></tr><tr><td>C++</td><td><code>.cpp</code>, <code>.cc</code>, <code>.cxx</code>, <code>.hpp</code>, <code>.hh</code>, <code>.hxx</code>, <code>.h</code>*</td><td>function, class, method, type, constant</td></tr><tr><td>Elixir</td><td><code>.ex</code>, <code>.exs</code></td><td>class (module/impl), type (protocol/@type/@callback), method, function</td></tr><tr><td>Ruby</td><td><code>.rb</code>, <code>.rake</code></td><td>class, type (module), method, function</td></tr><tr><td>SQL</td><td><code>.sql</code></td><td>function (CREATE FUNCTION, CTE, dbt macro/test/materialization), type (CREATE TABLE/VIEW/SCHEMA/INDEX, dbt snapshot)</td></tr><tr><td>XML/XUL</td><td><code>.xml</code>, <code>.xul</code></td><td>type (root element), constant (id attributes), function (script refs)</td></tr></tbody></table>
<p>* <code>.h</code> is parsed as C++ first, then falls back to C when no C++ symbols are extracted.</p>
<p>See LANGUAGE_SUPPORT.md for full semantics.</p>
<hr>
<h2>Context Providers</h2>
<p>When indexing local folders, jCodeMunch automatically detects ecosystem tools and enriches the index with <strong>business context</strong> — descriptions, tags, and metadata from project configuration files.</p>
<table><thead><tr><th>Provider</th><th>Detects</th><th>Enriches With</th></tr></thead><tbody><tr><td>dbt</td><td><code>dbt_project.yml</code></td><td>Model descriptions, tags, column names/descriptions</td></tr></tbody></table>
<p>Context enrichment is <strong>automatic</strong> — no configuration needed. When a provider detects its tool, it injects metadata into AI summarization prompts, file summaries, and search keywords.</p>
<p>Example: a dbt model with a <code>schema.yml</code> description produces file summaries like:</p>
<pre><code>This table summarizes account ledger. Tags: nightly, agg, intraday. 70 properties
</code></pre>
<p>Instead of the default:</p>
<pre><code>Contains 2 functions: source, renamed
</code></pre>
<p>The provider system is extensible — adding support for Terraform, OpenAPI, Django, or any other tool requires implementing a single <code>ContextProvider</code> class.</p>
<p>See CONTEXT_PROVIDERS.md for the full architecture, dbt details, and guide to writing new providers.</p>
<hr>
<h2>Contributing</h2>
<p>PRs welcome! All contributors must sign the <a href="https://cla-assistant.io/jgravelle/jcodemunch-mcp">Contributor License Agreement</a> before their PR can be merged — CLA Assistant will prompt you automatically. See <a href="CONTRIBUTING.md">CONTRIBUTING.md</a> for details.</p>
<hr>
<h2>Security</h2>
<p>Built-in protections:</p>
<ul>
<li>Path traversal prevention (owner/name sanitization + <code>_safe_content_path</code> enforcement)</li>
<li>Symlink escape protection</li>
<li>Secret file exclusion (<code>.env</code>, <code>*.pem</code>, etc.)</li>
<li>Binary detection</li>
<li>Configurable file size limits</li>
</ul>
<p>See SECURITY.md for details.</p>
<hr>
<h2>Best Use Cases</h2>
<ul>
<li>Large multi-module repositories</li>
<li>Agent-driven refactors</li>
<li>Architecture exploration</li>
<li>Faster onboarding</li>
<li>Token-efficient multi-agent workflows</li>
</ul>
<hr>
<h2>Not Intended For</h2>
<ul>
<li>LSP diagnostics or completions</li>
<li>Editing workflows</li>
<li>Real-time file watching</li>
<li>Cross-repository global indexing</li>
<li>Semantic program analysis</li>
</ul>
<hr>
<h2>Local LLMs (Ollama / LM Studio)</h2>
<p>You can use local, privacy-preserving AI models to generate summaries by providing an OpenAI-compatible endpoint.</p>
<p>For <strong>Ollama</strong>, run a model locally, then configure the MCP server:</p>
<pre><code class="language-json">"env": {
"OPENAI_API_BASE": "http://localhost:11434/v1",
"OPENAI_MODEL": "qwen3-coder"
}
</code></pre>
<p>For <strong>LM Studio</strong>, ensure the Local Server is running (usually on port 1234):</p>
<pre><code class="language-json">"env": {
"OPENAI_API_BASE": "http://127.0.0.1:1234/v1",
"OPENAI_MODEL": "openai/gpt-oss-20b"
}
</code></pre>
<blockquote>
<p>[!TIP]
<strong>Performance Note:</strong> Local models can be slow to load into memory on their first request, potentially causing the MCP server to time out and fall back to generic signature summaries. It is highly recommended to <strong>pre-load the model</strong> in Ollama or LM Studio before starting the server, or increase the <code>OPENAI_TIMEOUT</code> environment variable (e.g., to <code>"120.0"</code>) to allow more time for generation.</p>
</blockquote>
<hr>
<h2>Environment Variables</h2>
<table><thead><tr><th>Variable</th><th>Purpose</th><th>Required</th></tr></thead><tbody><tr><td><code>GITHUB_TOKEN</code></td><td>GitHub API auth</td><td>No</td></tr><tr><td><code>ANTHROPIC_API_KEY</code></td><td>Symbol summaries via Claude Haiku (takes priority)</td><td>No</td></tr><tr><td><code>ANTHROPIC_BASE_URL</code></td><td>Third-party Anthropic-compatible endpoints (e.g. z.ai)</td><td>No</td></tr><tr><td><code>ANTHROPIC_MODEL</code></td><td>Model name for Claude summaries (default: <code>claude-haiku-4-5-20251001</code>)</td><td>No</td></tr><tr><td><code>GOOGLE_API_KEY</code></td><td>Symbol summaries via Gemini Flash</td><td>No</td></tr><tr><td><code>GOOGLE_MODEL</code></td><td>Model name for Gemini summaries (default: <code>gemini-2.5-flash-lite</code>)</td><td>No</td></tr><tr><td><code>OPENAI_API_BASE</code></td><td>Base URL for local LLMs (e.g. <code>http://localhost:11434/v1</code>)</td><td>No</td></tr><tr><td><code>OPENAI_API_KEY</code></td><td>API key for local LLMs (default: <code>local-llm</code>)</td><td>No</td></tr><tr><td><code>OPENAI_MODEL</code></td><td>Model name for local LLMs (default: <code>qwen3-coder</code>)</td><td>No</td></tr><tr><td><code>OPENAI_TIMEOUT</code></td><td>Timeout in seconds for local requests (default: <code>60.0</code>)</td><td>No</td></tr><tr><td><code>OPENAI_BATCH_SIZE</code></td><td>Symbols per summarization request (default: <code>10</code>)</td><td>No</td></tr><tr><td><code>OPENAI_CONCURRENCY</code></td><td>Max parallel batch requests (default: <code>1</code>)</td><td>No</td></tr><tr><td><code>OPENAI_MAX_TOKENS</code></td><td>Max output tokens per batch response (default: <code>500</code>)</td><td>No</td></tr><tr><td><code>CODE_INDEX_PATH</code></td><td>Custom cache path</td><td>No</td></tr><tr><td><code>JCODEMUNCH_MAX_INDEX_FILES</code></td><td>Maximum files to index per repo/folder (default: <code>10000</code>)</td><td>No</td></tr><tr><td><code>JCODEMUNCH_CONTEXT_PROVIDERS</code></td><td>Set to <code>0</code> to disable context providers (dbt, etc.) during indexing</td><td>No</td></tr><tr><td><code>JCODEMUNCH_SHARE_SAVINGS</code></td><td>Set to <code>0</code> to disable anonymous community token savings reporting</td><td>No</td></tr><tr><td><code>JCODEMUNCH_LOG_LEVEL</code></td><td>Log level: <code>DEBUG</code>, <code>INFO</code>, <code>WARNING</code>, <code>ERROR</code> (default: <code>WARNING</code>)</td><td>No</td></tr><tr><td><code>JCODEMUNCH_LOG_FILE</code></td><td>Path to log file. If unset, logs go to stderr. Use a file to avoid polluting MCP stdio.</td><td>No</td></tr></tbody></table>
<h3>Community Savings Meter</h3>
<p>Each tool call contributes an anonymous delta to a live global counter at <a href="https://j.gravelle.us">j.gravelle.us</a>. Only two values are ever sent: the tokens saved (a number) and a random anonymous install ID — never code, paths, repo names, or anything identifying. The anon ID is generated once and stored in <code>~/.code-index/_savings.json</code>.</p>
<p>To disable, set <code>JCODEMUNCH_SHARE_SAVINGS=0</code> in your MCP server env.</p>
<hr>
<h2>Documentation</h2>
<ul>
<li>USER_GUIDE.md</li>
<li>ARCHITECTURE.md</li>
<li>SPEC.md</li>
<li>SECURITY.md</li>
<li>LANGUAGE_SUPPORT.md</li>
<li>CONTEXT_PROVIDERS.md</li>
</ul>
<hr>
<h2>Star History</h2>
<!-- --><a href="https://www.star-history.com/#jgravelle/jcodemunch-mcp&type=date&legend=top-left">
<picture>
<source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=jgravelle/jcodemunch-mcp&type=date&theme=dark&legend=top-left" />
<source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/svg?repos=jgravelle/jcodemunch-mcp&type=date&legend=top-left" />
<img alt="Star History Chart" src="https://api.star-history.com/svg?repos=jgravelle/jcodemunch-mcp&type=date&legend=top-left" />
</picture>
</a><!-- -->
<hr>
<h2>License (Dual Use)</h2>
<p>This repository is <strong>free for non-commercial use</strong> under the terms below.<br>
<strong>Commercial use requires a paid commercial license.</strong></p>
<hr>
<h2>Copyright and License Text</h2>
<p>Copyright (c) 2026 J. Gravelle</p>
<h3>1. Non-Commercial License Grant (Free)</h3>
<p>Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to use, copy, modify, merge, publish, and distribute the Software for <strong>personal, educational, research, hobby, or other non-commercial purposes</strong>, subject to the following conditions:</p>
<ol>
<li>
<p>The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.</p>
</li>
<li>
<p>Any modifications made to the Software must clearly indicate that they are derived from the original work, and the name of the original author (J. Gravelle) must remain intact. He's kinda full of himself.</p>
</li>
<li>
<p>Redistributions of the Software in source code form must include a prominent notice describing any modifications from the original version.</p>
</li>
</ol>
<h3>2. Commercial Use</h3>
<p>Commercial use of the Software requires a separate paid commercial license from the author.</p>
<p>“Commercial use” includes, but is not limited to:</p>
<ul>
<li>Use of the Software in a business environment</li>
<li>Internal use within a for-profit organization</li>
<li>Incorporation into a product or service offered for sale</li>
<li>Use in connection with revenue generation, consulting, SaaS, hosting, or fee-based services</li>
</ul>
<p>For commercial licensing inquiries, contact:<br>
<a href="/cdn-cgi/l/email-protection#89e3c9eefbe8ffece5e5eca7fcfa"><span class="__cf_email__" data-cfemail="b5dff5d2c7d4c3d0d9d9d09bc0c6">[email protected]</span></a> | <a href="https://j.gravelle.us">https://j.gravelle.us</a></p>
<p>Until a commercial license is obtained, commercial use is not permitted.</p>
<h3>3. Disclaimer of Warranty</h3>
<p>THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NONINFRINGEMENT.</p>
<p>IN NO EVENT SHALL THE AUTHOR OR COPYRIGHT HOLDER BE LIABLE FOR ANY CLAIM, DAMAGES, OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT, OR OTHERWISE, ARISING FROM, OUT OF, OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.</p> | https://mcpservers.org/servers/jgravelle/jcodemunch-mcp | https://mcpservers.org/all?sort=newest&page=1 |
harvest-mcp-server | Harvest time tracking integration with 40+ tools for managing time entries, projects, clients, tasks, and generating time reports via the Harvest API v2 | https://github.com/ianaleck/harvest-mcp-server | 🌾 Harvest MCP Server | 🌾 Harvest MCP Server Unofficial Model Context Protocol (MCP) server for seamless integration with the Harvest time tracking API ⚠️ Disclaimer: This is an unofficial, third-party integration with the Harvest API. This project is not affiliated with, endorsed by, or sponsored by Harvest or Forecast (the company behind Harvest). ✨ Features 🔗 Complete Harvest API v2 Coverage - 40+ tools covering all major endpoints 🛡️ Type-Safe - Full TypeScript support with Zod validation ⚡ High Performance - Built with async/await and proper rate limiting 🧪 Thoroughly Tested - Comprehensive unit, integration, and contract tests 📊 Rich Logging - Structured logging for debugging and monitoring 🔄 Auto-Retry - Intelligent retry logic with exponential backoff 📖 MCP Compliant - Works with Claude Desktop and other MCP clients 🚀 Quick Start Prerequisites Node.js 18+ Harvest account with API access MCP-compatible client (like Claude Desktop) Installation # Install globally npm install -g @ianaleck/harvest-mcp-server # Or install locally npm install @ianaleck/harvest-mcp-server Configuration Get your Harvest API credentials: Go to Harvest → Settings → Developers → Personal Access Tokens Create a new token Note your Account ID (visible in URL or settings) Configure your MCP client (e.g., Claude Desktop): { "mcpServers": { "harvest": { "command": "npx", "args": ["-y", "@ianaleck/harvest-mcp-server"], "env": { "HARVEST_ACCESS_TOKEN": "your_harvest_personal_access_token", "HARVEST_ACCOUNT_ID": "your_harvest_account_id" } } } } Start using with Claude! 🎯 What You Can Do Once connected, you can ask Claude to help with: ⏱️ Time Tracking "Show me all my time entries for this week" "Start a timer for the 'Development' task on the 'Website Project'" "How many hours did I work on Project X last month?" 📋 Project Management "List all active projects for client Acme Corp" "Create a new project called 'Mobile App' for client TechStart" "Show me project budget vs actual time spent" 👥 Team Management "Who are all the users in our Harvest account?" "Show me John's time entries for last week" 💰 Financial Tracking "Generate an expense report for Q4" "Show me all unpaid invoices" "What's our total billable hours this month?" 🛠️ Available Tools <details> <summary><strong>📊 Company & Account (1 tool)</strong></summary> get_company - Get company information and settings </details> <details> <summary><strong>⏰ Time Entries (8 tools)</strong></summary> list_time_entries - List time entries with filtering get_time_entry - Get specific time entry details create_time_entry - Create new time entry update_time_entry - Update existing time entry delete_time_entry - Delete time entry start_timer - Start a timer for a task stop_timer - Stop running timer restart_timer - Restart a previous time entry </details> <details> <summary><strong>🏗️ Projects (7 tools)</strong></summary> list_projects - List all projects with filtering get_project - Get specific project details create_project - Create new project update_project - Update project details delete_project - Delete project list_project_task_assignments - List task assignments for project create_project_task_assignment - Assign task to project update_project_task_assignment - Update task assignment delete_project_task_assignment - Remove task assignment </details> <details> <summary><strong>📝 Tasks (5 tools)</strong></summary> list_tasks - List all tasks get_task - Get specific task details create_task - Create new task update_task - Update task details delete_task - Delete task </details> <details> <summary><strong>🏢 Clients (5 tools)</strong></summary> list_clients - List all clients get_client - Get specific client details create_client - Create new client update_client - Update client details delete_client - Delete client </details> <details> <summary><strong>👤 Users (6 tools)</strong></summary> list_users - List all users in account get_user - Get specific user details get_current_user - Get current authenticated user create_user - Create new user update_user - Update user details delete_user - Delete user </details> <details> <summary><strong>💸 Expenses (6 tools)</strong></summary> list_expenses - List expenses with filtering get_expense - Get specific expense details create_expense - Create new expense update_expense - Update expense details delete_expense - Delete expense list_expense_categories - List all expense categories </details> <details> <summary><strong>🧾 Invoices (5 tools)</strong></summary> list_invoices - List invoices with filtering get_invoice - Get specific invoice details create_invoice - Create new invoice update_invoice - Update invoice details delete_invoice - Delete invoice </details> <details> <summary><strong>📋 Estimates (5 tools)</strong></summary> list_estimates - List estimates with filtering get_estimate - Get specific estimate details create_estimate - Create new estimate update_estimate - Update estimate details delete_estimate - Delete estimate </details> <details> <summary><strong>📈 Reports (4 tools)</strong></summary> get_time_report - Generate time reports with filtering get_expense_report - Generate expense reports get_project_budget_report - Get project budget analysis get_uninvoiced_report - Get uninvoiced time and expenses </details> 🧪 Development Setup git clone https://github.com/ianaleck/harvest-mcp-server.git cd harvest-mcp-server npm install Environment Configuration cp .env.example .env # Edit .env with your Harvest API credentials Testing # Run all tests npm test # Run with coverage npm run test:coverage # Run specific test suites npm run test:unit npm run test:integration npm run test:contract Building # Build for production npm run build # Start development server npm run dev 📋 API Requirements This server requires a Harvest account with API access. Users must comply with: Harvest API Terms of Service Harvest API Rate Limits (100 requests per 15 seconds) 🤝 Contributing We welcome contributions! Please see our Contributing Guidelines for details. Fork the repository Create a feature branch ( git checkout -b feature/amazing-feature ) Make your changes with tests Ensure all tests pass ( npm test ) Commit your changes ( git commit -m 'Add amazing feature' ) Push to the branch ( git push origin feature/amazing-feature ) Open a Pull Request 📄 License This project is licensed under the MIT License - see the LICENSE file for details. 🙏 Acknowledgments Harvest for providing an excellent time tracking API Model Context Protocol team for the MCP specification Anthropic for Claude and the MCP SDK 📞 Support 🐛 Bug Reports: GitHub Issues 💡 Feature Requests: GitHub Discussions 📖 Documentation: MCP Documentation <div align="center"> Made with ❤️ for the MCP community ⭐ Star this project if you find it useful! </div> | <h1>🌾 Harvest MCP Server</h1>
<p><a href="https://www.npmjs.com/package/@ianaleck/harvest-mcp-server"><img alt="npm version" src="https://badge.fury.io/js/@ianaleck%2Fharvest-mcp-server.svg"></a>
<a href="https://www.typescriptlang.org/"><img alt="TypeScript" src="https://img.shields.io/badge/TypeScript-007ACC?logo=typescript&logoColor=white"></a>
<a href="https://modelcontextprotocol.io/"><img alt="MCP" src="https://img.shields.io/badge/MCP-Compatible-blue"></a>
<a href="https://opensource.org/licenses/MIT"><img alt="License: MIT" src="https://img.shields.io/badge/License-MIT-yellow.svg"></a>
<a href="https://github.com/ianaleck/harvest-mcp-server"><img alt="Tests" src="https://img.shields.io/badge/Tests-Passing-green"></a>
<a href="https://buymeacoffee.com/ianaleck"><img alt="Buy Me A Coffee" src="https://img.shields.io/badge/Buy%20Me%20A%20Coffee-FFDD00?logo=buy-me-a-coffee&logoColor=black"></a></p>
<blockquote>
<p><strong>Unofficial</strong> Model Context Protocol (MCP) server for seamless integration with the Harvest time tracking API</p>
</blockquote>
<p><strong>⚠️ Disclaimer:</strong> This is an unofficial, third-party integration with the Harvest API. This project is not affiliated with, endorsed by, or sponsored by Harvest or Forecast (the company behind Harvest).</p>
<p><a href="https://glama.ai/mcp/servers/ianaleck/harvest-mcp-server"><img alt="harvest-mcp-server MCP server" src="https://glama.ai/mcp/servers/ianaleck/harvest-mcp-server/badges/card.svg"></a></p>
<h2>✨ Features</h2>
<ul>
<li>🔗 <strong>Complete Harvest API v2 Coverage</strong> - 40+ tools covering all major endpoints</li>
<li>🛡️ <strong>Type-Safe</strong> - Full TypeScript support with Zod validation</li>
<li>⚡ <strong>High Performance</strong> - Built with async/await and proper rate limiting</li>
<li>🧪 <strong>Thoroughly Tested</strong> - Comprehensive unit, integration, and contract tests</li>
<li>📊 <strong>Rich Logging</strong> - Structured logging for debugging and monitoring</li>
<li>🔄 <strong>Auto-Retry</strong> - Intelligent retry logic with exponential backoff</li>
<li>📖 <strong>MCP Compliant</strong> - Works with Claude Desktop and other MCP clients</li>
</ul>
<h2>🚀 Quick Start</h2>
<h3>Prerequisites</h3>
<ul>
<li>Node.js 18+</li>
<li>Harvest account with API access</li>
<li>MCP-compatible client (like Claude Desktop)</li>
</ul>
<h3>Installation</h3>
<pre><code class="language-bash"># Install globally
npm install -g @ianaleck/harvest-mcp-server
# Or install locally
npm install @ianaleck/harvest-mcp-server
</code></pre>
<h3>Configuration</h3>
<ol>
<li>
<p><strong>Get your Harvest API credentials:</strong></p>
<ul>
<li>Go to Harvest → Settings → Developers → Personal Access Tokens</li>
<li>Create a new token</li>
<li>Note your Account ID (visible in URL or settings)</li>
</ul>
</li>
<li>
<p><strong>Configure your MCP client</strong> (e.g., Claude Desktop):</p>
</li>
</ol>
<pre><code class="language-json">{
"mcpServers": {
"harvest": {
"command": "npx",
"args": ["-y", "@ianaleck/harvest-mcp-server"],
"env": {
"HARVEST_ACCESS_TOKEN": "your_harvest_personal_access_token",
"HARVEST_ACCOUNT_ID": "your_harvest_account_id"
}
}
}
}
</code></pre>
<ol start="3">
<li><strong>Start using with Claude!</strong></li>
</ol>
<h2>🎯 What You Can Do</h2>
<p>Once connected, you can ask Claude to help with:</p>
<h3>⏱️ Time Tracking</h3>
<ul>
<li>"Show me all my time entries for this week"</li>
<li>"Start a timer for the 'Development' task on the 'Website Project'"</li>
<li>"How many hours did I work on Project X last month?"</li>
</ul>
<h3>📋 Project Management</h3>
<ul>
<li>"List all active projects for client Acme Corp"</li>
<li>"Create a new project called 'Mobile App' for client TechStart"</li>
<li>"Show me project budget vs actual time spent"</li>
</ul>
<h3>👥 Team Management</h3>
<ul>
<li>"Who are all the users in our Harvest account?"</li>
<li>"Show me John's time entries for last week"</li>
</ul>
<h3>💰 Financial Tracking</h3>
<ul>
<li>"Generate an expense report for Q4"</li>
<li>"Show me all unpaid invoices"</li>
<li>"What's our total billable hours this month?"</li>
</ul>
<h2>🛠️ Available Tools</h2>
<!-- --><details>
<summary><strong>📊 Company & Account (1 tool)</strong></summary><!-- -->
<ul>
<li><code>get_company</code> - Get company information and settings</li>
</ul>
<!-- --></details><!-- -->
<!-- --><details>
<summary><strong>⏰ Time Entries (8 tools)</strong></summary><!-- -->
<ul>
<li><code>list_time_entries</code> - List time entries with filtering</li>
<li><code>get_time_entry</code> - Get specific time entry details</li>
<li><code>create_time_entry</code> - Create new time entry</li>
<li><code>update_time_entry</code> - Update existing time entry</li>
<li><code>delete_time_entry</code> - Delete time entry</li>
<li><code>start_timer</code> - Start a timer for a task</li>
<li><code>stop_timer</code> - Stop running timer</li>
<li><code>restart_timer</code> - Restart a previous time entry</li>
</ul>
<!-- --></details><!-- -->
<!-- --><details>
<summary><strong>🏗️ Projects (7 tools)</strong></summary><!-- -->
<ul>
<li><code>list_projects</code> - List all projects with filtering</li>
<li><code>get_project</code> - Get specific project details</li>
<li><code>create_project</code> - Create new project</li>
<li><code>update_project</code> - Update project details</li>
<li><code>delete_project</code> - Delete project</li>
<li><code>list_project_task_assignments</code> - List task assignments for project</li>
<li><code>create_project_task_assignment</code> - Assign task to project</li>
<li><code>update_project_task_assignment</code> - Update task assignment</li>
<li><code>delete_project_task_assignment</code> - Remove task assignment</li>
</ul>
<!-- --></details><!-- -->
<!-- --><details>
<summary><strong>📝 Tasks (5 tools)</strong></summary><!-- -->
<ul>
<li><code>list_tasks</code> - List all tasks</li>
<li><code>get_task</code> - Get specific task details</li>
<li><code>create_task</code> - Create new task</li>
<li><code>update_task</code> - Update task details</li>
<li><code>delete_task</code> - Delete task</li>
</ul>
<!-- --></details><!-- -->
<!-- --><details>
<summary><strong>🏢 Clients (5 tools)</strong></summary><!-- -->
<ul>
<li><code>list_clients</code> - List all clients</li>
<li><code>get_client</code> - Get specific client details</li>
<li><code>create_client</code> - Create new client</li>
<li><code>update_client</code> - Update client details</li>
<li><code>delete_client</code> - Delete client</li>
</ul>
<!-- --></details><!-- -->
<!-- --><details>
<summary><strong>👤 Users (6 tools)</strong></summary><!-- -->
<ul>
<li><code>list_users</code> - List all users in account</li>
<li><code>get_user</code> - Get specific user details</li>
<li><code>get_current_user</code> - Get current authenticated user</li>
<li><code>create_user</code> - Create new user</li>
<li><code>update_user</code> - Update user details</li>
<li><code>delete_user</code> - Delete user</li>
</ul>
<!-- --></details><!-- -->
<!-- --><details>
<summary><strong>💸 Expenses (6 tools)</strong></summary><!-- -->
<ul>
<li><code>list_expenses</code> - List expenses with filtering</li>
<li><code>get_expense</code> - Get specific expense details</li>
<li><code>create_expense</code> - Create new expense</li>
<li><code>update_expense</code> - Update expense details</li>
<li><code>delete_expense</code> - Delete expense</li>
<li><code>list_expense_categories</code> - List all expense categories</li>
</ul>
<!-- --></details><!-- -->
<!-- --><details>
<summary><strong>🧾 Invoices (5 tools)</strong></summary><!-- -->
<ul>
<li><code>list_invoices</code> - List invoices with filtering</li>
<li><code>get_invoice</code> - Get specific invoice details</li>
<li><code>create_invoice</code> - Create new invoice</li>
<li><code>update_invoice</code> - Update invoice details</li>
<li><code>delete_invoice</code> - Delete invoice</li>
</ul>
<!-- --></details><!-- -->
<!-- --><details>
<summary><strong>📋 Estimates (5 tools)</strong></summary><!-- -->
<ul>
<li><code>list_estimates</code> - List estimates with filtering</li>
<li><code>get_estimate</code> - Get specific estimate details</li>
<li><code>create_estimate</code> - Create new estimate</li>
<li><code>update_estimate</code> - Update estimate details</li>
<li><code>delete_estimate</code> - Delete estimate</li>
</ul>
<!-- --></details><!-- -->
<!-- --><details>
<summary><strong>📈 Reports (4 tools)</strong></summary><!-- -->
<ul>
<li><code>get_time_report</code> - Generate time reports with filtering</li>
<li><code>get_expense_report</code> - Generate expense reports</li>
<li><code>get_project_budget_report</code> - Get project budget analysis</li>
<li><code>get_uninvoiced_report</code> - Get uninvoiced time and expenses</li>
</ul>
<!-- --></details><!-- -->
<h2>🧪 Development</h2>
<h3>Setup</h3>
<pre><code class="language-bash">git clone https://github.com/ianaleck/harvest-mcp-server.git
cd harvest-mcp-server
npm install
</code></pre>
<h3>Environment Configuration</h3>
<pre><code class="language-bash">cp .env.example .env
# Edit .env with your Harvest API credentials
</code></pre>
<h3>Testing</h3>
<pre><code class="language-bash"># Run all tests
npm test
# Run with coverage
npm run test:coverage
# Run specific test suites
npm run test:unit
npm run test:integration
npm run test:contract
</code></pre>
<h3>Building</h3>
<pre><code class="language-bash"># Build for production
npm run build
# Start development server
npm run dev
</code></pre>
<h2>📋 API Requirements</h2>
<p>This server requires a Harvest account with API access. Users must comply with:</p>
<ul>
<li><a href="https://help.getharvest.com/api-v2/introduction/overview/general/">Harvest API Terms of Service</a></li>
<li><a href="https://help.getharvest.com/api-v2/introduction/overview/general/#rate-limiting">Harvest API Rate Limits</a> (100 requests per 15 seconds)</li>
</ul>
<h2>🤝 Contributing</h2>
<p>We welcome contributions! Please see our <a href="CONTRIBUTING.md">Contributing Guidelines</a> for details.</p>
<ol>
<li>Fork the repository</li>
<li>Create a feature branch (<code>git checkout -b feature/amazing-feature</code>)</li>
<li>Make your changes with tests</li>
<li>Ensure all tests pass (<code>npm test</code>)</li>
<li>Commit your changes (<code>git commit -m 'Add amazing feature'</code>)</li>
<li>Push to the branch (<code>git push origin feature/amazing-feature</code>)</li>
<li>Open a Pull Request</li>
</ol>
<h2>📄 License</h2>
<p>This project is licensed under the MIT License - see the <a href="LICENSE">LICENSE</a> file for details.</p>
<h2>🙏 Acknowledgments</h2>
<ul>
<li><a href="https://www.getharvest.com/">Harvest</a> for providing an excellent time tracking API</li>
<li><a href="https://modelcontextprotocol.io/">Model Context Protocol</a> team for the MCP specification</li>
<li><a href="https://www.anthropic.com/">Anthropic</a> for Claude and the MCP SDK</li>
</ul>
<h2>📞 Support</h2>
<ul>
<li>🐛 <strong>Bug Reports:</strong> <a href="https://github.com/ianaleck/harvest-mcp-server/issues">GitHub Issues</a></li>
<li>💡 <strong>Feature Requests:</strong> <a href="https://github.com/ianaleck/harvest-mcp-server/discussions">GitHub Discussions</a></li>
<li>📖 <strong>Documentation:</strong> <a href="https://modelcontextprotocol.io/docs">MCP Documentation</a></li>
</ul>
<hr>
<!-- --><div align="center"><!-- -->
<p><strong>Made with ❤️ for the MCP community</strong></p>
<p><a href="https://github.com/ianaleck/harvest-mcp-server">⭐ Star this project</a> if you find it useful!</p>
<!-- --></div> | https://mcpservers.org/servers/ianaleck/harvest-mcp-server | https://mcpservers.org/all?sort=newest&page=1 |
Fastmail MCP Server | An open-source server that gives AI assistants full access to Fastmail email, calendars, and contacts over JMAP. | https://github.com/Jordonh18/fastmail-mcp-server | Fastmail MCP Server | Fastmail MCP Server A Model Context Protocol (MCP) server for Fastmail. Enables any MCP-compatible AI assistant to read, search, send, and manage emails, calendars, and contacts in a Fastmail account via the JMAP protocol. Prerequisites Node.js 18 or later A Fastmail account (Standard or Professional plan) A Fastmail API token with JMAP access Setup 1. Generate a Fastmail API Token Log in to Fastmail Go to Settings → Privacy & Security → Manage API tokens Click New API token Grant access to Mail , Calendars , and Contacts (read and write) Copy the generated token 2. Install npm install @jordonh19/fastmail-mcp-server Or run directly: npx @jordonh19/fastmail-mcp-server 3. Configure This MCP server works with any AI assistant that supports the Model Context Protocol. Below are examples for popular MCP clients. Claude Desktop Add to your Claude Desktop configuration ( claude_desktop_config.json ): { "mcpServers": { "fastmail": { "command": "npx", "args": ["@jordonh19/fastmail-mcp-server"], "env": { "FASTMAIL_API_TOKEN": "your-api-token-here" } } } } Claude Code CLI export FASTMAIL_API_TOKEN="your-api-token-here" claude mcp add fastmail -- npx @jordonh19/fastmail-mcp-server Cursor Add to your Cursor MCP configuration ( .cursor/mcp.json ): { "mcpServers": { "fastmail": { "command": "npx", "args": ["@jordonh19/fastmail-mcp-server"], "env": { "FASTMAIL_API_TOKEN": "your-api-token-here" } } } } Windsurf Add to your Windsurf MCP configuration ( ~/.codeium/windsurf/mcp_config.json ): { "mcpServers": { "fastmail": { "command": "npx", "args": ["@jordonh19/fastmail-mcp-server"], "env": { "FASTMAIL_API_TOKEN": "your-api-token-here" } } } } VS Code Add to your VS Code MCP configuration ( .vscode/mcp.json ): { "servers": { "fastmail": { "command": "npx", "args": ["@jordonh19/fastmail-mcp-server"], "env": { "FASTMAIL_API_TOKEN": "your-api-token-here" } } } } Other MCP Clients For any MCP-compatible client, run the server with the FASTMAIL_API_TOKEN environment variable set: FASTMAIL_API_TOKEN="your-api-token-here" npx @jordonh19/fastmail-mcp-server The server communicates over stdio by default, following the standard MCP transport protocol. To run over HTTP instead, use the --transport flag: FASTMAIL_API_TOKEN="your-api-token-here" npx @jordonh19/fastmail-mcp-server --transport http --port 3000 Tools Email search_emails : Search emails by mailbox, text, sender, date range, attachments, and more. get_email : Get full email content by ID. get_thread : Get all emails in a conversation thread. get_unread_emails : Quickly retrieve unread emails, optionally filtered by mailbox. get_latest_emails : Get the most recent emails from all or a specific mailbox. get_mailbox_emails : List emails in a specific mailbox with pagination. get_email_attachments : List attachments on an email without returning the full message body. send_email : Compose and send a new email. reply_email : Reply or reply-all to an email. forward_email : Forward an email to new recipients. create_draft : Save an email as a draft without sending. send_draft : Send a previously saved draft email. move_email : Move an email to a different mailbox. add_labels : Add mailbox labels to an email while preserving existing mailbox assignments. remove_labels : Remove mailbox labels from an email while preserving other mailbox assignments. update_email_flags : Mark emails as read/unread or flagged/unflagged. delete_email : Move to Trash or permanently delete. bulk_email_action : Perform actions on multiple emails at once. bulk_add_labels : Add mailbox labels to multiple emails at once. bulk_remove_labels : Remove mailbox labels from multiple emails at once. archive_email : Move one or more emails to the Archive mailbox. mark_mailbox_read : Mark all emails in a mailbox as read. get_mailbox_stats : Get compact mailbox-level unread, email, and thread counts. get_account_summary : Get a compact account overview with unique email totals and top mailboxes. download_attachment : Download an email attachment by blob ID. Mailbox list_mailboxes : List all mailboxes or folders with roles and email counts. create_mailbox : Create a new mailbox or folder. rename_mailbox : Rename an existing mailbox or folder. delete_mailbox : Delete a mailbox or folder, optionally with force delete. Calendar list_calendars : List all calendars with names, colors, and visibility. get_calendar_events : Search or list calendar events by date range, calendar, or title. get_calendar_event : Get full details of a specific calendar event. create_calendar_event : Create a new calendar event with location, participants, and alerts. update_calendar_event : Update an existing calendar event. delete_calendar_event : Delete a calendar event. Contacts list_address_books : List all address books or contact groups. search_contacts : Search contacts by name, email, or other criteria. get_contact : Get full details of a specific contact. create_contact : Create a new contact with email, phone, organization, and more. update_contact : Update an existing contact's information. delete_contact : Delete a contact. Identity get_identities : List available sender identities. Diagnostics check_function_availability : Check which Fastmail feature groups and MCP client capabilities are available, with setup guidance for missing access. Transport Modes The server supports two transport modes: stdio (default) Standard input/output transport. Used by most MCP clients (Claude Desktop, Cursor, VS Code, etc.): FASTMAIL_API_TOKEN="your-token" npx @jordonh19/fastmail-mcp-server HTTP (Streamable HTTP) Runs as an HTTP server for remote access or multi-client scenarios: FASTMAIL_API_TOKEN="your-token" npx @jordonh19/fastmail-mcp-server --transport http --port 3000 The HTTP transport exposes a single /mcp endpoint that supports the MCP Streamable HTTP protocol. Web UI Dashboard When running in HTTP mode, a built-in web dashboard is available at the server root. It provides: Live tool-call log — see every MCP tool invocation in real time via SSE Connection tracking — monitor active MCP client connections Server uptime — at-a-glance health status On startup the server prints a one-time access token to the console. Open http://localhost:<port>/ in a browser and enter the token to log in. The token is bound to an HttpOnly cookie so it never appears in URLs. Claude Desktop Extension (DXT) A pre-packaged .dxt extension can be built for one-click installation in Claude Desktop: npm run build:dxt This produces fastmail-mcp-server-v<version>.dxt in the project root. Double-click the file (or drag it into Claude Desktop) to install. Claude will prompt for your Fastmail API token on first use. Configuration The server can be configured via environment variables or a JSON config file. Environment Variables FASTMAIL_API_TOKEN : Required. Fastmail API token with JMAP access. Config File Create a .fastmail-mcp.json file in your project root or home directory: { "transport": "stdio", "port": 3000 } The server searches for config files in this order: ./.fastmail-mcp.json (current directory) ~/.fastmail-mcp.json (home directory) Environment variables and CLI flags always take precedence over config file values. Development See CONTRIBUTING.md for detailed development guidelines. git clone https://github.com/Jordonh18/fastmail-mcp-server.git cd fastmail-mcp-server npm install npm run build npm test Scripts Command Description npm run build TypeScript compilation npm run build:dxt Build + package as a Claude Desktop .dxt extension npm run dev Watch mode build npm test Run tests (Vitest) npm run test:coverage Tests with v8 coverage npm run typecheck Type check without emitting Run locally: FASTMAIL_API_TOKEN="your-token" node dist/index.js License MIT | <h1>Fastmail MCP Server</h1>
<p>A Model Context Protocol (MCP) server for Fastmail. Enables any MCP-compatible AI assistant to read, search, send, and manage emails, calendars, and contacts in a Fastmail account via the JMAP protocol.</p>
<h2>Prerequisites</h2>
<ul>
<li>Node.js 18 or later</li>
<li>A Fastmail account (Standard or Professional plan)</li>
<li>A Fastmail API token with JMAP access</li>
</ul>
<h2>Setup</h2>
<h3>1. Generate a Fastmail API Token</h3>
<ol>
<li>Log in to <a href="https://www.fastmail.com">Fastmail</a></li>
<li>Go to <strong>Settings → Privacy & Security → Manage API tokens</strong></li>
<li>Click <strong>New API token</strong></li>
<li>Grant access to <strong>Mail</strong>, <strong>Calendars</strong>, and <strong>Contacts</strong> (read and write)</li>
<li>Copy the generated token</li>
</ol>
<h3>2. Install</h3>
<pre><code class="language-bash">npm install @jordonh19/fastmail-mcp-server
</code></pre>
<p>Or run directly:</p>
<pre><code class="language-bash">npx @jordonh19/fastmail-mcp-server
</code></pre>
<h3>3. Configure</h3>
<p>This MCP server works with any AI assistant that supports the Model Context Protocol. Below are examples for popular MCP clients.</p>
<h4>Claude Desktop</h4>
<p>Add to your Claude Desktop configuration (<code>claude_desktop_config.json</code>):</p>
<pre><code class="language-json">{
"mcpServers": {
"fastmail": {
"command": "npx",
"args": ["@jordonh19/fastmail-mcp-server"],
"env": {
"FASTMAIL_API_TOKEN": "your-api-token-here"
}
}
}
}
</code></pre>
<h4>Claude Code CLI</h4>
<pre><code class="language-bash">export FASTMAIL_API_TOKEN="your-api-token-here"
claude mcp add fastmail -- npx @jordonh19/fastmail-mcp-server
</code></pre>
<h4>Cursor</h4>
<p>Add to your Cursor MCP configuration (<code>.cursor/mcp.json</code>):</p>
<pre><code class="language-json">{
"mcpServers": {
"fastmail": {
"command": "npx",
"args": ["@jordonh19/fastmail-mcp-server"],
"env": {
"FASTMAIL_API_TOKEN": "your-api-token-here"
}
}
}
}
</code></pre>
<h4>Windsurf</h4>
<p>Add to your Windsurf MCP configuration (<code>~/.codeium/windsurf/mcp_config.json</code>):</p>
<pre><code class="language-json">{
"mcpServers": {
"fastmail": {
"command": "npx",
"args": ["@jordonh19/fastmail-mcp-server"],
"env": {
"FASTMAIL_API_TOKEN": "your-api-token-here"
}
}
}
}
</code></pre>
<h4>VS Code</h4>
<p>Add to your VS Code MCP configuration (<code>.vscode/mcp.json</code>):</p>
<pre><code class="language-json">{
"servers": {
"fastmail": {
"command": "npx",
"args": ["@jordonh19/fastmail-mcp-server"],
"env": {
"FASTMAIL_API_TOKEN": "your-api-token-here"
}
}
}
}
</code></pre>
<h4>Other MCP Clients</h4>
<p>For any MCP-compatible client, run the server with the <code>FASTMAIL_API_TOKEN</code> environment variable set:</p>
<pre><code class="language-bash">FASTMAIL_API_TOKEN="your-api-token-here" npx @jordonh19/fastmail-mcp-server
</code></pre>
<p>The server communicates over stdio by default, following the standard MCP transport protocol. To run over HTTP instead, use the <code>--transport</code> flag:</p>
<pre><code class="language-bash">FASTMAIL_API_TOKEN="your-api-token-here" npx @jordonh19/fastmail-mcp-server --transport http --port 3000
</code></pre>
<h2>Tools</h2>
<h3>Email</h3>
<ul>
<li><code>search_emails</code>: Search emails by mailbox, text, sender, date range, attachments, and more.</li>
<li><code>get_email</code>: Get full email content by ID.</li>
<li><code>get_thread</code>: Get all emails in a conversation thread.</li>
<li><code>get_unread_emails</code>: Quickly retrieve unread emails, optionally filtered by mailbox.</li>
<li><code>get_latest_emails</code>: Get the most recent emails from all or a specific mailbox.</li>
<li><code>get_mailbox_emails</code>: List emails in a specific mailbox with pagination.</li>
<li><code>get_email_attachments</code>: List attachments on an email without returning the full message body.</li>
<li><code>send_email</code>: Compose and send a new email.</li>
<li><code>reply_email</code>: Reply or reply-all to an email.</li>
<li><code>forward_email</code>: Forward an email to new recipients.</li>
<li><code>create_draft</code>: Save an email as a draft without sending.</li>
<li><code>send_draft</code>: Send a previously saved draft email.</li>
<li><code>move_email</code>: Move an email to a different mailbox.</li>
<li><code>add_labels</code>: Add mailbox labels to an email while preserving existing mailbox assignments.</li>
<li><code>remove_labels</code>: Remove mailbox labels from an email while preserving other mailbox assignments.</li>
<li><code>update_email_flags</code>: Mark emails as read/unread or flagged/unflagged.</li>
<li><code>delete_email</code>: Move to Trash or permanently delete.</li>
<li><code>bulk_email_action</code>: Perform actions on multiple emails at once.</li>
<li><code>bulk_add_labels</code>: Add mailbox labels to multiple emails at once.</li>
<li><code>bulk_remove_labels</code>: Remove mailbox labels from multiple emails at once.</li>
<li><code>archive_email</code>: Move one or more emails to the Archive mailbox.</li>
<li><code>mark_mailbox_read</code>: Mark all emails in a mailbox as read.</li>
<li><code>get_mailbox_stats</code>: Get compact mailbox-level unread, email, and thread counts.</li>
<li><code>get_account_summary</code>: Get a compact account overview with unique email totals and top mailboxes.</li>
<li><code>download_attachment</code>: Download an email attachment by blob ID.</li>
</ul>
<h3>Mailbox</h3>
<ul>
<li><code>list_mailboxes</code>: List all mailboxes or folders with roles and email counts.</li>
<li><code>create_mailbox</code>: Create a new mailbox or folder.</li>
<li><code>rename_mailbox</code>: Rename an existing mailbox or folder.</li>
<li><code>delete_mailbox</code>: Delete a mailbox or folder, optionally with force delete.</li>
</ul>
<h3>Calendar</h3>
<ul>
<li><code>list_calendars</code>: List all calendars with names, colors, and visibility.</li>
<li><code>get_calendar_events</code>: Search or list calendar events by date range, calendar, or title.</li>
<li><code>get_calendar_event</code>: Get full details of a specific calendar event.</li>
<li><code>create_calendar_event</code>: Create a new calendar event with location, participants, and alerts.</li>
<li><code>update_calendar_event</code>: Update an existing calendar event.</li>
<li><code>delete_calendar_event</code>: Delete a calendar event.</li>
</ul>
<h3>Contacts</h3>
<ul>
<li><code>list_address_books</code>: List all address books or contact groups.</li>
<li><code>search_contacts</code>: Search contacts by name, email, or other criteria.</li>
<li><code>get_contact</code>: Get full details of a specific contact.</li>
<li><code>create_contact</code>: Create a new contact with email, phone, organization, and more.</li>
<li><code>update_contact</code>: Update an existing contact's information.</li>
<li><code>delete_contact</code>: Delete a contact.</li>
</ul>
<h3>Identity</h3>
<ul>
<li><code>get_identities</code>: List available sender identities.</li>
</ul>
<h3>Diagnostics</h3>
<ul>
<li><code>check_function_availability</code>: Check which Fastmail feature groups and MCP client capabilities are available, with setup guidance for missing access.</li>
</ul>
<h2>Transport Modes</h2>
<p>The server supports two transport modes:</p>
<h3>stdio (default)</h3>
<p>Standard input/output transport. Used by most MCP clients (Claude Desktop, Cursor, VS Code, etc.):</p>
<pre><code class="language-bash">FASTMAIL_API_TOKEN="your-token" npx @jordonh19/fastmail-mcp-server
</code></pre>
<h3>HTTP (Streamable HTTP)</h3>
<p>Runs as an HTTP server for remote access or multi-client scenarios:</p>
<pre><code class="language-bash">FASTMAIL_API_TOKEN="your-token" npx @jordonh19/fastmail-mcp-server --transport http --port 3000
</code></pre>
<p>The HTTP transport exposes a single <code>/mcp</code> endpoint that supports the MCP Streamable HTTP protocol.</p>
<h3>Web UI Dashboard</h3>
<p>When running in HTTP mode, a built-in web dashboard is available at the server root. It provides:</p>
<ul>
<li><strong>Live tool-call log</strong> — see every MCP tool invocation in real time via SSE</li>
<li><strong>Connection tracking</strong> — monitor active MCP client connections</li>
<li><strong>Server uptime</strong> — at-a-glance health status</li>
</ul>
<p>On startup the server prints a one-time access token to the console. Open <code>http://localhost:<port>/</code> in a browser and enter the token to log in. The token is bound to an HttpOnly cookie so it never appears in URLs.</p>
<h2>Claude Desktop Extension (DXT)</h2>
<p>A pre-packaged <code>.dxt</code> extension can be built for one-click installation in Claude Desktop:</p>
<pre><code class="language-bash">npm run build:dxt
</code></pre>
<p>This produces <code>fastmail-mcp-server-v<version>.dxt</code> in the project root. Double-click the file (or drag it into Claude Desktop) to install. Claude will prompt for your Fastmail API token on first use.</p>
<h2>Configuration</h2>
<p>The server can be configured via environment variables or a JSON config file.</p>
<h3>Environment Variables</h3>
<ul>
<li><code>FASTMAIL_API_TOKEN</code>: Required. Fastmail API token with JMAP access.</li>
</ul>
<h3>Config File</h3>
<p>Create a <code>.fastmail-mcp.json</code> file in your project root or home directory:</p>
<pre><code class="language-json">{
"transport": "stdio",
"port": 3000
}
</code></pre>
<p>The server searches for config files in this order:</p>
<ol>
<li><code>./.fastmail-mcp.json</code> (current directory)</li>
<li><code>~/.fastmail-mcp.json</code> (home directory)</li>
</ol>
<p>Environment variables and CLI flags always take precedence over config file values.</p>
<h2>Development</h2>
<p>See <a href="CONTRIBUTING.md">CONTRIBUTING.md</a> for detailed development guidelines.</p>
<pre><code class="language-bash">git clone https://github.com/Jordonh18/fastmail-mcp-server.git
cd fastmail-mcp-server
npm install
npm run build
npm test
</code></pre>
<h3>Scripts</h3>
<table><thead><tr><th>Command</th><th>Description</th></tr></thead><tbody><tr><td><code>npm run build</code></td><td>TypeScript compilation</td></tr><tr><td><code>npm run build:dxt</code></td><td>Build + package as a Claude Desktop <code>.dxt</code> extension</td></tr><tr><td><code>npm run dev</code></td><td>Watch mode build</td></tr><tr><td><code>npm test</code></td><td>Run tests (Vitest)</td></tr><tr><td><code>npm run test:coverage</code></td><td>Tests with v8 coverage</td></tr><tr><td><code>npm run typecheck</code></td><td>Type check without emitting</td></tr></tbody></table>
<p>Run locally:</p>
<pre><code class="language-bash">FASTMAIL_API_TOKEN="your-token" node dist/index.js
</code></pre>
<h2>License</h2>
<p>MIT</p> | https://mcpservers.org/servers/jordonh18/fastmail-mcp-server | https://mcpservers.org/all?sort=newest&page=1 |
jDocMunch-MCP | jDocMunch-MCP lets AI agents navigate documentation by section instead of reading files by brute force. | https://github.com/jgravelle/jdocmunch-mcp | jDocMunch MCP | Stop Feeding Documentation Trees to Your AI Most AI agents still explore documentation the expensive way: open file → skim hundreds of irrelevant paragraphs → open another file → repeat That burns tokens, floods context windows with noise, and forces models to reason through a lot of text they never needed in the first place. jDocMunch-MCP lets AI agents navigate documentation by section instead of reading files by brute force. It indexes a documentation set once, then retrieves exactly the section the agent actually needs, with byte-precise extraction from the original file. Task Traditional approach With jDocMunch Find a configuration section ~12,000 tokens ~400 tokens Browse documentation structure ~40,000 tokens ~800 tokens Explore a full doc set ~100,000 tokens ~2,000 tokens Index once. Query cheaply forever. Precision context beats brute-force context. jDocMunch MCP AI-native documentation navigation for serious agents Commercial licenses jDocMunch-MCP is free for non-commercial use . Commercial use requires a paid license. jDocMunch-only licenses Builder — $29 — 1 developer Studio — $99 — up to 5 developers Platform — $499 — org-wide internal deployment Want both code and docs retrieval? Munch Duo Builder Bundle — $89 Munch Duo Studio Bundle — $399 Munch Duo Platform Bundle — $2,249 Stop dumping documentation files into context windows. Start navigating docs structurally. jDocMunch indexes documentation once by heading hierarchy and section structure, then gives MCP-compatible agents precise access to the explanations they actually need instead of forcing them to brute-read files. It is built for workflows where token efficiency, context hygiene, and agent reliability matter. Why this exists Large context windows do not fix bad retrieval. Agents waste money and reasoning bandwidth when they: open entire documents to find one configuration block repeatedly re-read headings, boilerplate, and unrelated sections lose important explanations inside oversized context payloads consume documentation as flat text instead of structured knowledge jDocMunch fixes that by changing the unit of access from file to section . Instead of handing an agent an entire document, it can retrieve exactly: an installation section a configuration section an API explanation a troubleshooting section a specific subtree of related headings That makes documentation exploration cheaper, faster, and more stable. What makes it different Section-first retrieval Search and retrieve documentation by section, not just file path or keyword match. Byte-precise extraction Full content is pulled on demand from exact byte offsets into the original file. Stable section IDs Sections retain durable identities across re-indexing when path, heading text, and heading level remain unchanged. Local-first architecture Indexes and raw docs are stored locally. No hosted dependency required. MCP-native workflow Works with Claude Desktop, Claude Code, Google Antigravity, and other MCP-compatible clients. What gets indexed Every section stores: title and heading level one-line summary extracted tags and references SHA-256 content hash for drift detection byte offsets into the original file This allows agents to discover documentation structurally, then request only the specific section they need. Why agents need this Traditional doc retrieval methods all break in different ways: File scanning loads far too much irrelevant text Keyword search finds terms but often loses context Chunking breaks authored hierarchy and separates explanations from examples jDocMunch preserves the structure the human author intended: heading hierarchy parent/child relationships section boundaries coherent explanatory units Agents do not need bigger context windows. They need better navigation. How it works jDocMunch implements jMRI-Full — the open specification for structured retrieval MCP servers. jMRI-Full covers the full stack: discover, search, retrieve, and metadata operations with batch retrieval, hash-based drift detection, byte-offset addressing, and a complete _meta envelope on every call. Discovery GitHub API or local directory walk Security filtering Traversal protection, secret exclusion, binary detection Parsing Format-aware section splitting: heading-based (Markdown/MDX/HTML/RST/AsciiDoc), structure-based (OpenAPI tags, JSON keys, XML elements), or cell-based (Jupyter) Hierarchy wiring Parent/child relationships established Summarization Heading text → AI batch summaries → title fallback Storage JSON index + raw files stored locally under ~/.doc-index/ Retrieval O(1) byte-offset seeking via stable section IDs Stable section IDs {repo}::{doc_path}::{ancestor-chain/slug}#{level} The slug is prefixed with the ancestor heading chain, making IDs both readable and stable. A new heading inserted in one branch of a document never renumbers IDs in another branch. Examples: owner/repo::docs/install.md::installation#1 owner/repo::docs/install.md::installation/prerequisites#3 owner/repo::README.md::usage/configuration/advanced-configuration#4 local/myproject::guide.md::configuration#2 IDs remain stable across re-indexing when the file path, heading text, heading level, and parent heading chain do not change. Installation Prerequisites Python 3.10+ pip Install pip install jdocmunch-mcp Verify: jdocmunch-mcp --help Configure an MCP client PATH note: MCP clients often run with a restricted environment where jdocmunch-mcp may not be found even if it works in your shell. Using uvx is the recommended approach because it resolves the package on demand without relying on your system PATH. If you prefer pip install , use the absolute path to the executable instead. Common executable paths Linux: /home/<username>/.local/bin/jdocmunch-mcp macOS: /Users/<username>/.local/bin/jdocmunch-mcp Windows: C:\\Users\\<username>\\AppData\\Roaming\\Python\\Python3xx\\Scripts\\jdocmunch-mcp.exe Claude Desktop / Claude Code Config file location: OS Path macOS ~/Library/Application Support/Claude/claude_desktop_config.json Linux ~/.config/claude/claude_desktop_config.json Windows %APPDATA%\Claude\claude_desktop_config.json Minimal config { "mcpServers": { "jdocmunch": { "command": "uvx", "args": ["jdocmunch-mcp"] } } } With optional AI summaries and GitHub auth { "mcpServers": { "jdocmunch": { "command": "uvx", "args": ["jdocmunch-mcp"], "env": { "GITHUB_TOKEN": "ghp_...", "ANTHROPIC_API_KEY": "sk-ant-..." } } } } After saving the config, restart Claude Desktop / Claude Code . Google Antigravity Open the Agent pane Click the ⋯ menu → MCP Servers → Manage MCP Servers Click View raw config to open mcp_config.json Add the entry below, save, then restart the MCP server { "mcpServers": { "jdocmunch": { "command": "uvx", "args": ["jdocmunch-mcp"] } } } Usage examples index_local: { "path": "/path/to/docs" } index_repo: { "url": "owner/repo" } get_toc: { "repo": "owner/repo" } get_toc_tree: { "repo": "owner/repo" } get_document_outline: { "repo": "owner/repo", "doc_path": "docs/config.md" } search_sections: { "repo": "owner/repo", "query": "authentication" } get_section: { "repo": "owner/repo", "section_id": "owner/repo::docs/config.md::authentication#1" } Tool surface Tool Purpose index_local Index a local documentation folder index_repo Index a GitHub repository’s docs list_repos List indexed documentation sets get_toc Flat section list in document order get_toc_tree Nested section tree per document get_document_outline Section hierarchy for one document search_sections Weighted search returning summaries only get_section Full content of one section get_sections Batch content retrieval get_section_context Section + ancestor headings + child summaries delete_index Remove a doc index Search and retrieval tools include a _meta envelope with timing, token savings, and cost avoided. Example: "_meta": { "latency_ms": 12, "sections_returned": 5, "tokens_saved": 1840, "total_tokens_saved": 94320, "cost_avoided": { "claude_opus": 0.0276, "gpt5_latest": 0.0184 }, "total_cost_avoided": { "claude_opus": 1.4148, "gpt5_latest": 0.9432 } } total_tokens_saved and total_cost_avoided accumulate across tool calls and persist to ~/.doc-index/_savings.json . Supported formats Format Extensions Notes Markdown .md , .markdown ATX ( # Heading ) and setext headings MDX .mdx JSX tags, frontmatter, import/export stripped before parsing Plain text .txt Paragraph-block section splitting reStructuredText .rst Adornment-based heading detection AsciiDoc .adoc = and == heading hierarchy Jupyter Notebook .ipynb Markdown cells used as sections; code cells attached as content HTML .html <h1> – <h6> headings; boilerplate stripped OpenAPI / Swagger .yaml , .yml , .json , .jsonc OpenAPI 3.x and Swagger 2.x; operations grouped by tag as sections JSON / JSONC .json , .jsonc Top-level keys as sections; JSONC comments stripped before parsing XML / SVG / XHTML .xml , .svg , .xhtml Element hierarchy used for section structure See ARCHITECTURE.md for parser details. Security Built-in protections include: path traversal prevention symlink escape protection secret file exclusion ( .env , *.pem , and similar) binary file detection configurable file size limits storage path injection prevention via _safe_content_path() atomic index writes See SECURITY.md for details. Best use cases agent-driven documentation exploration finding configuration and API reference sections onboarding to unfamiliar frameworks token-efficient multi-agent documentation workflows large documentation sets with dozens of files Not intended for source code symbol indexing (use jCodeMunch for that) real-time file watching cross-repository global search semantic/vector similarity search as a standalone product (semantic search is supported as an enhancement when embeddings are enabled via use_embeddings=true , but the core workflow is structure-first) Environment variables Variable Purpose Required GITHUB_TOKEN GitHub API auth No ANTHROPIC_API_KEY Section summaries via Claude Haiku No GOOGLE_API_KEY Section summaries via Gemini Flash; also Gemini embeddings No OPENAI_API_KEY OpenAI embeddings (text-embedding-3-small) No JDOCMUNCH_EMBEDDING_PROVIDER Force provider: gemini , openai , sentence-transformers , none No JDOCMUNCH_ST_MODEL sentence-transformers model (default: all-MiniLM-L6-v2 ) No DOC_INDEX_PATH Custom cache path No JDOCMUNCH_SHARE_SAVINGS Set to 0 to disable anonymous community token savings reporting No Community savings meter Each tool call can contribute an anonymous delta to a live global counter at j.gravelle.us . Only two values are sent: tokens saved a random anonymous install ID No content, file paths, repo names, or identifying material are sent. The anonymous install ID is generated once and stored in ~/.doc-index/_savings.json . To disable reporting, set: JDOCMUNCH_SHARE_SAVINGS=0 Contributing PRs welcome! All contributors must sign the Contributor License Agreement before their PR can be merged — CLA Assistant will prompt you automatically. See CONTRIBUTING.md for details. Documentation USER_GUIDE.md ARCHITECTURE.md SPEC.md SECURITY.md TOKEN_SAVINGS.md License (dual use) This repository is free for non-commercial use under the terms below. Commercial use requires a paid commercial license. Star History <a href="https://www.star-history.com/?repos=jgravelle%2Fjdocmunch-mcp&type=date&legend=top-left"> <picture> <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/image?repos=jgravelle/jdocmunch-mcp&type=date&theme=dark&legend=top-left" /> <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/image?repos=jgravelle/jdocmunch-mcp&type=date&legend=top-left" /> <img alt="Star History Chart" src="https://api.star-history.com/image?repos=jgravelle/jdocmunch-mcp&type=date&legend=top-left" /> </picture> </a> Copyright and license text Copyright (c) 2026 J. Gravelle 1. Non-commercial license grant (free) Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to use, copy, modify, merge, publish, and distribute the Software for personal, educational, research, hobby, or other non-commercial purposes , subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. Any modifications made to the Software must clearly indicate that they are derived from the original work, and the name of the original author (J. Gravelle) must remain intact. He's kinda full of himself. Redistributions of the Software in source code form must include a prominent notice describing any modifications from the original version. 2. Commercial use Commercial use of the Software requires a separate paid commercial license from the author. “Commercial use” includes, but is not limited to: use of the Software in a business environment internal use within a for-profit organization incorporation into a product or service offered for sale use in connection with revenue generation, consulting, SaaS, hosting, or fee-based services For commercial licensing inquiries: [email protected] https://j.gravelle.us Until a commercial license is obtained, commercial use is not permitted. 3. Disclaimer of warranty THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHOR OR COPYRIGHT HOLDER BE LIABLE FOR ANY CLAIM, DAMAGES, OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT, OR OTHERWISE, ARISING FROM, OUT OF, OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. | <h2>Stop Feeding Documentation Trees to Your AI</h2>
<p>Most AI agents still explore documentation the expensive way:</p>
<p>open file → skim hundreds of irrelevant paragraphs → open another file → repeat</p>
<p>That burns tokens, floods context windows with noise, and forces models to reason through a lot of text they never needed in the first place.</p>
<p><strong>jDocMunch-MCP lets AI agents navigate documentation by section instead of reading files by brute force.</strong><br>
<!-- -->It indexes a documentation set once, then retrieves exactly the section the agent actually needs, with byte-precise extraction from the original file.</p>
<table><thead><tr><th>Task</th><th style="text-align:right">Traditional approach</th><th style="text-align:right">With jDocMunch</th></tr></thead><tbody><tr><td>Find a configuration section</td><td style="text-align:right">~12,000 tokens</td><td style="text-align:right">~400 tokens</td></tr><tr><td>Browse documentation structure</td><td style="text-align:right">~40,000 tokens</td><td style="text-align:right">~800 tokens</td></tr><tr><td>Explore a full doc set</td><td style="text-align:right">~100,000 tokens</td><td style="text-align:right">~2,000 tokens</td></tr></tbody></table>
<p>Index once. Query cheaply forever.<br>
<strong>Precision context beats brute-force context.</strong></p>
<hr>
<h1>jDocMunch MCP</h1>
<h3>AI-native documentation navigation for serious agents</h3>
<p><img alt="License" src="https://img.shields.io/badge/license-dual--use-blue">
<img alt="MCP" src="https://img.shields.io/badge/MCP-compatible-purple">
<img alt="Local-first" src="https://img.shields.io/badge/local--first-yes-brightgreen">
<img alt="jMRI" src="https://img.shields.io/badge/jMRI-Full-blueviolet">
<a href="https://pypi.org/project/jdocmunch-mcp/"><img alt="PyPI version" src="https://img.shields.io/pypi/v/jdocmunch-mcp"></a>
<a href="https://pypi.org/project/jdocmunch-mcp/"><img alt="PyPI - Python Version" src="https://img.shields.io/pypi/pyversions/jdocmunch-mcp"></a></p>
<blockquote>
<h2>Commercial licenses</h2>
<p>jDocMunch-MCP is <strong>free for non-commercial use</strong>.</p>
<p><strong>Commercial use requires a paid license.</strong></p>
<p><strong>jDocMunch-only licenses</strong></p>
<ul>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#builder">Builder — $29</a> — 1 developer</li>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#studio">Studio — $99</a> — up to 5 developers</li>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#platform">Platform — $499</a> — org-wide internal deployment</li>
</ul>
<p><strong>Want both code and docs retrieval?</strong></p>
<ul>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#builder">Munch Duo Builder Bundle — $89</a></li>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#studio">Munch Duo Studio Bundle — $399</a></li>
<li><a href="https://j.gravelle.us/jCodeMunch/descriptions.php#platform">Munch Duo Platform Bundle — $2,249</a></li>
</ul>
</blockquote>
<p><strong>Stop dumping documentation files into context windows. Start navigating docs structurally.</strong></p>
<p>jDocMunch indexes documentation once by heading hierarchy and section structure, then gives MCP-compatible agents precise access to the explanations they actually need instead of forcing them to brute-read files.</p>
<p>It is built for workflows where token efficiency, context hygiene, and agent reliability matter.</p>
<hr>
<h2>Why this exists</h2>
<p>Large context windows do not fix bad retrieval.</p>
<p>Agents waste money and reasoning bandwidth when they:</p>
<ul>
<li>open entire documents to find one configuration block</li>
<li>repeatedly re-read headings, boilerplate, and unrelated sections</li>
<li>lose important explanations inside oversized context payloads</li>
<li>consume documentation as flat text instead of structured knowledge</li>
</ul>
<p>jDocMunch fixes that by changing the unit of access from <strong>file</strong> to <strong>section</strong>.</p>
<p>Instead of handing an agent an entire document, it can retrieve exactly:</p>
<ul>
<li>an installation section</li>
<li>a configuration section</li>
<li>an API explanation</li>
<li>a troubleshooting section</li>
<li>a specific subtree of related headings</li>
</ul>
<p>That makes documentation exploration cheaper, faster, and more stable.</p>
<hr>
<h2>What makes it different</h2>
<h3>Section-first retrieval</h3>
<p>Search and retrieve documentation by section, not just file path or keyword match.</p>
<h3>Byte-precise extraction</h3>
<p>Full content is pulled on demand from exact byte offsets into the original file.</p>
<h3>Stable section IDs</h3>
<p>Sections retain durable identities across re-indexing when path, heading text, and heading level remain unchanged.</p>
<h3>Local-first architecture</h3>
<p>Indexes and raw docs are stored locally. No hosted dependency required.</p>
<h3>MCP-native workflow</h3>
<p>Works with Claude Desktop, Claude Code, Google Antigravity, and other MCP-compatible clients.</p>
<hr>
<h2>What gets indexed</h2>
<p>Every section stores:</p>
<ul>
<li>title and heading level</li>
<li>one-line summary</li>
<li>extracted tags and references</li>
<li>SHA-256 content hash for drift detection</li>
<li>byte offsets into the original file</li>
</ul>
<p>This allows agents to discover documentation structurally, then request only the specific section they need.</p>
<hr>
<h2>Why agents need this</h2>
<p>Traditional doc retrieval methods all break in different ways:</p>
<ul>
<li><strong>File scanning</strong> loads far too much irrelevant text</li>
<li><strong>Keyword search</strong> finds terms but often loses context</li>
<li><strong>Chunking</strong> breaks authored hierarchy and separates explanations from examples</li>
</ul>
<p>jDocMunch preserves the structure the human author intended:</p>
<ul>
<li>heading hierarchy</li>
<li>parent/child relationships</li>
<li>section boundaries</li>
<li>coherent explanatory units</li>
</ul>
<p>Agents do not need bigger context windows.<br>
<!-- -->They need better navigation.</p>
<hr>
<h2>How it works</h2>
<p>jDocMunch implements <strong><a href="https://dev.to/jgravelle/your-ai-agent-is-dumpster-diving-through-your-code-326f">jMRI-Full</a></strong> — the open specification for structured retrieval MCP servers. jMRI-Full covers the full stack: discover, search, retrieve, and metadata operations with batch retrieval, hash-based drift detection, byte-offset addressing, and a complete <code>_meta</code> envelope on every call.</p>
<ol>
<li>
<p><strong>Discovery</strong>
GitHub API or local directory walk</p>
</li>
<li>
<p><strong>Security filtering</strong>
Traversal protection, secret exclusion, binary detection</p>
</li>
<li>
<p><strong>Parsing</strong>
Format-aware section splitting: heading-based (Markdown/MDX/HTML/RST/AsciiDoc), structure-based (OpenAPI tags, JSON keys, XML elements), or cell-based (Jupyter)</p>
</li>
<li>
<p><strong>Hierarchy wiring</strong>
Parent/child relationships established</p>
</li>
<li>
<p><strong>Summarization</strong>
Heading text → AI batch summaries → title fallback</p>
</li>
<li>
<p><strong>Storage</strong>
JSON index + raw files stored locally under <code>~/.doc-index/</code></p>
</li>
<li>
<p><strong>Retrieval</strong>
O(1) byte-offset seeking via stable section IDs</p>
</li>
</ol>
<hr>
<h2>Stable section IDs</h2>
<pre><code class="language-text">{repo}::{doc_path}::{ancestor-chain/slug}#{level}
</code></pre>
<p>The slug is prefixed with the ancestor heading chain, making IDs both readable and stable. A new heading inserted in one branch of a document never renumbers IDs in another branch.</p>
<p>Examples:</p>
<ul>
<li><code>owner/repo::docs/install.md::installation#1</code></li>
<li><code>owner/repo::docs/install.md::installation/prerequisites#3</code></li>
<li><code>owner/repo::README.md::usage/configuration/advanced-configuration#4</code></li>
<li><code>local/myproject::guide.md::configuration#2</code></li>
</ul>
<p>IDs remain stable across re-indexing when the file path, heading text, heading level, and parent heading chain do not change.</p>
<hr>
<h2>Installation</h2>
<h3>Prerequisites</h3>
<ul>
<li>Python 3.10+</li>
<li><code>pip</code></li>
</ul>
<h3>Install</h3>
<pre><code class="language-bash">pip install jdocmunch-mcp
</code></pre>
<p>Verify:</p>
<pre><code class="language-bash">jdocmunch-mcp --help
</code></pre>
<hr>
<h2>Configure an MCP client</h2>
<blockquote>
<p><strong>PATH note:</strong> MCP clients often run with a restricted environment where <code>jdocmunch-mcp</code> may not be found even if it works in your shell. Using <a href="https://github.com/astral-sh/uv"><code>uvx</code></a> is the recommended approach because it resolves the package on demand without relying on your system PATH. If you prefer <code>pip install</code>, use the absolute path to the executable instead.</p>
</blockquote>
<h3>Common executable paths</h3>
<ul>
<li><strong>Linux:</strong> <code>/home/<username>/.local/bin/jdocmunch-mcp</code></li>
<li><strong>macOS:</strong> <code>/Users/<username>/.local/bin/jdocmunch-mcp</code></li>
<li><strong>Windows:</strong> <code>C:\\Users\\<username>\\AppData\\Roaming\\Python\\Python3xx\\Scripts\\jdocmunch-mcp.exe</code></li>
</ul>
<hr>
<h2>Claude Desktop / Claude Code</h2>
<p>Config file location:</p>
<table><thead><tr><th>OS</th><th>Path</th></tr></thead><tbody><tr><td>macOS</td><td><code>~/Library/Application Support/Claude/claude_desktop_config.json</code></td></tr><tr><td>Linux</td><td><code>~/.config/claude/claude_desktop_config.json</code></td></tr><tr><td>Windows</td><td><code>%APPDATA%\Claude\claude_desktop_config.json</code></td></tr></tbody></table>
<h3>Minimal config</h3>
<pre><code class="language-json">{
"mcpServers": {
"jdocmunch": {
"command": "uvx",
"args": ["jdocmunch-mcp"]
}
}
}
</code></pre>
<h3>With optional AI summaries and GitHub auth</h3>
<pre><code class="language-json">{
"mcpServers": {
"jdocmunch": {
"command": "uvx",
"args": ["jdocmunch-mcp"],
"env": {
"GITHUB_TOKEN": "ghp_...",
"ANTHROPIC_API_KEY": "sk-ant-..."
}
}
}
}
</code></pre>
<p>After saving the config, <strong>restart Claude Desktop / Claude Code</strong>.</p>
<hr>
<h2>Google Antigravity</h2>
<ol>
<li>Open the Agent pane</li>
<li>Click the <code>⋯</code> menu → <strong>MCP Servers</strong> → <strong>Manage MCP Servers</strong></li>
<li>Click <strong>View raw config</strong> to open <code>mcp_config.json</code></li>
<li>Add the entry below, save, then restart the MCP server</li>
</ol>
<pre><code class="language-json">{
"mcpServers": {
"jdocmunch": {
"command": "uvx",
"args": ["jdocmunch-mcp"]
}
}
}
</code></pre>
<hr>
<h2>Usage examples</h2>
<pre><code class="language-json">index_local: { "path": "/path/to/docs" }
index_repo: { "url": "owner/repo" }
get_toc: { "repo": "owner/repo" }
get_toc_tree: { "repo": "owner/repo" }
get_document_outline: { "repo": "owner/repo", "doc_path": "docs/config.md" }
search_sections: { "repo": "owner/repo", "query": "authentication" }
get_section: { "repo": "owner/repo", "section_id": "owner/repo::docs/config.md::authentication#1" }
</code></pre>
<hr>
<h2>Tool surface</h2>
<table><thead><tr><th>Tool</th><th>Purpose</th></tr></thead><tbody><tr><td><code>index_local</code></td><td>Index a local documentation folder</td></tr><tr><td><code>index_repo</code></td><td>Index a GitHub repository’s docs</td></tr><tr><td><code>list_repos</code></td><td>List indexed documentation sets</td></tr><tr><td><code>get_toc</code></td><td>Flat section list in document order</td></tr><tr><td><code>get_toc_tree</code></td><td>Nested section tree per document</td></tr><tr><td><code>get_document_outline</code></td><td>Section hierarchy for one document</td></tr><tr><td><code>search_sections</code></td><td>Weighted search returning summaries only</td></tr><tr><td><code>get_section</code></td><td>Full content of one section</td></tr><tr><td><code>get_sections</code></td><td>Batch content retrieval</td></tr><tr><td><code>get_section_context</code></td><td>Section + ancestor headings + child summaries</td></tr><tr><td><code>delete_index</code></td><td>Remove a doc index</td></tr></tbody></table>
<p>Search and retrieval tools include a <code>_meta</code> envelope with timing, token savings, and cost avoided.</p>
<p>Example:</p>
<pre><code class="language-json">"_meta": {
"latency_ms": 12,
"sections_returned": 5,
"tokens_saved": 1840,
"total_tokens_saved": 94320,
"cost_avoided": { "claude_opus": 0.0276, "gpt5_latest": 0.0184 },
"total_cost_avoided": { "claude_opus": 1.4148, "gpt5_latest": 0.9432 }
}
</code></pre>
<p><code>total_tokens_saved</code> and <code>total_cost_avoided</code> accumulate across tool calls and persist to <code>~/.doc-index/_savings.json</code>.</p>
<hr>
<h2>Supported formats</h2>
<table><thead><tr><th>Format</th><th>Extensions</th><th>Notes</th></tr></thead><tbody><tr><td>Markdown</td><td><code>.md</code>, <code>.markdown</code></td><td>ATX (<code># Heading</code>) and setext headings</td></tr><tr><td>MDX</td><td><code>.mdx</code></td><td>JSX tags, frontmatter, import/export stripped before parsing</td></tr><tr><td>Plain text</td><td><code>.txt</code></td><td>Paragraph-block section splitting</td></tr><tr><td>reStructuredText</td><td><code>.rst</code></td><td>Adornment-based heading detection</td></tr><tr><td>AsciiDoc</td><td><code>.adoc</code></td><td><code>=</code> and <code>==</code> heading hierarchy</td></tr><tr><td>Jupyter Notebook</td><td><code>.ipynb</code></td><td>Markdown cells used as sections; code cells attached as content</td></tr><tr><td>HTML</td><td><code>.html</code></td><td><code><h1></code>–<code><h6></code> headings; boilerplate stripped</td></tr><tr><td>OpenAPI / Swagger</td><td><code>.yaml</code>, <code>.yml</code>, <code>.json</code>, <code>.jsonc</code></td><td>OpenAPI 3.x and Swagger 2.x; operations grouped by tag as sections</td></tr><tr><td>JSON / JSONC</td><td><code>.json</code>, <code>.jsonc</code></td><td>Top-level keys as sections; JSONC comments stripped before parsing</td></tr><tr><td>XML / SVG / XHTML</td><td><code>.xml</code>, <code>.svg</code>, <code>.xhtml</code></td><td>Element hierarchy used for section structure</td></tr></tbody></table>
<p>See <code>ARCHITECTURE.md</code> for parser details.</p>
<hr>
<h2>Security</h2>
<p>Built-in protections include:</p>
<ul>
<li>path traversal prevention</li>
<li>symlink escape protection</li>
<li>secret file exclusion (<code>.env</code>, <code>*.pem</code>, and similar)</li>
<li>binary file detection</li>
<li>configurable file size limits</li>
<li>storage path injection prevention via <code>_safe_content_path()</code></li>
<li>atomic index writes</li>
</ul>
<p>See <code>SECURITY.md</code> for details.</p>
<hr>
<h2>Best use cases</h2>
<ul>
<li>agent-driven documentation exploration</li>
<li>finding configuration and API reference sections</li>
<li>onboarding to unfamiliar frameworks</li>
<li>token-efficient multi-agent documentation workflows</li>
<li>large documentation sets with dozens of files</li>
</ul>
<hr>
<h2>Not intended for</h2>
<ul>
<li>source code symbol indexing (use <a href="https://github.com/jgravelle/jcodemunch-mcp">jCodeMunch</a> for that)</li>
<li>real-time file watching</li>
<li>cross-repository global search</li>
<li>semantic/vector similarity search as a standalone product (semantic search is supported as an enhancement when embeddings are enabled via <code>use_embeddings=true</code>, but the core workflow is structure-first)</li>
</ul>
<hr>
<h2>Environment variables</h2>
<table><thead><tr><th>Variable</th><th>Purpose</th><th>Required</th></tr></thead><tbody><tr><td><code>GITHUB_TOKEN</code></td><td>GitHub API auth</td><td>No</td></tr><tr><td><code>ANTHROPIC_API_KEY</code></td><td>Section summaries via Claude Haiku</td><td>No</td></tr><tr><td><code>GOOGLE_API_KEY</code></td><td>Section summaries via Gemini Flash; also Gemini embeddings</td><td>No</td></tr><tr><td><code>OPENAI_API_KEY</code></td><td>OpenAI embeddings (text-embedding-3-small)</td><td>No</td></tr><tr><td><code>JDOCMUNCH_EMBEDDING_PROVIDER</code></td><td>Force provider: <code>gemini</code>, <code>openai</code>, <code>sentence-transformers</code>, <code>none</code></td><td>No</td></tr><tr><td><code>JDOCMUNCH_ST_MODEL</code></td><td>sentence-transformers model (default: <code>all-MiniLM-L6-v2</code>)</td><td>No</td></tr><tr><td><code>DOC_INDEX_PATH</code></td><td>Custom cache path</td><td>No</td></tr><tr><td><code>JDOCMUNCH_SHARE_SAVINGS</code></td><td>Set to <code>0</code> to disable anonymous community token savings reporting</td><td>No</td></tr></tbody></table>
<hr>
<h2>Community savings meter</h2>
<p>Each tool call can contribute an anonymous delta to a live global counter at <a href="https://j.gravelle.us">j.gravelle.us</a>. Only two values are sent:</p>
<ul>
<li>tokens saved</li>
<li>a random anonymous install ID</li>
</ul>
<p>No content, file paths, repo names, or identifying material are sent.</p>
<p>The anonymous install ID is generated once and stored in <code>~/.doc-index/_savings.json</code>.</p>
<p>To disable reporting, set:</p>
<pre><code class="language-bash">JDOCMUNCH_SHARE_SAVINGS=0
</code></pre>
<hr>
<h2>Contributing</h2>
<p>PRs welcome! All contributors must sign the <a href="https://cla-assistant.io/jgravelle/jdocmunch-mcp">Contributor License Agreement</a> before their PR can be merged — CLA Assistant will prompt you automatically. See <a href="CONTRIBUTING.md">CONTRIBUTING.md</a> for details.</p>
<hr>
<h2>Documentation</h2>
<ul>
<li><a href="USER_GUIDE.md">USER_GUIDE.md</a></li>
<li><a href="ARCHITECTURE.md">ARCHITECTURE.md</a></li>
<li><a href="SPEC.md">SPEC.md</a></li>
<li><a href="SECURITY.md">SECURITY.md</a></li>
<li><a href="TOKEN_SAVINGS.md">TOKEN_SAVINGS.md</a></li>
</ul>
<hr>
<h2>License (dual use)</h2>
<p>This repository is <strong>free for non-commercial use</strong> under the terms below.
<strong>Commercial use requires a paid commercial license.</strong></p>
<hr>
<h2>Star History</h2>
<!-- --><a href="https://www.star-history.com/?repos=jgravelle%2Fjdocmunch-mcp&type=date&legend=top-left">
<picture>
<source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/image?repos=jgravelle/jdocmunch-mcp&type=date&theme=dark&legend=top-left" />
<source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/image?repos=jgravelle/jdocmunch-mcp&type=date&legend=top-left" />
<img alt="Star History Chart" src="https://api.star-history.com/image?repos=jgravelle/jdocmunch-mcp&type=date&legend=top-left" />
</picture>
</a><!-- -->
<hr>
<h2>Copyright and license text</h2>
<p>Copyright (c) 2026 J. Gravelle</p>
<h3>1. Non-commercial license grant (free)</h3>
<p>Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to use, copy, modify, merge, publish, and distribute the Software for <strong>personal, educational, research, hobby, or other non-commercial purposes</strong>, subject to the following conditions:</p>
<ol>
<li>The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.</li>
<li>Any modifications made to the Software must clearly indicate that they are derived from the original work, and the name of the original author (J. Gravelle) must remain intact. He's kinda full of himself.</li>
<li>Redistributions of the Software in source code form must include a prominent notice describing any modifications from the original version.</li>
</ol>
<h3>2. Commercial use</h3>
<p>Commercial use of the Software requires a separate paid commercial license from the author.</p>
<p>“Commercial use” includes, but is not limited to:</p>
<ul>
<li>use of the Software in a business environment</li>
<li>internal use within a for-profit organization</li>
<li>incorporation into a product or service offered for sale</li>
<li>use in connection with revenue generation, consulting, SaaS, hosting, or fee-based services</li>
</ul>
<p>For commercial licensing inquiries:
<strong><a href="/cdn-cgi/l/email-protection#e78da780958691828b8b82c99294"><span class="__cf_email__" data-cfemail="e18ba186938097848d8d84cf9492">[email protected]</span></a></strong>
<strong><a href="https://j.gravelle.us">https://j.gravelle.us</a></strong></p>
<p>Until a commercial license is obtained, commercial use is not permitted.</p>
<h3>3. Disclaimer of warranty</h3>
<p>THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NONINFRINGEMENT.</p>
<p>IN NO EVENT SHALL THE AUTHOR OR COPYRIGHT HOLDER BE LIABLE FOR ANY CLAIM, DAMAGES, OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT, OR OTHERWISE, ARISING FROM, OUT OF, OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.</p> | https://mcpservers.org/servers/jgravelle/jdocmunch-mcp | https://mcpservers.org/all?sort=newest&page=1 |
End of preview. Expand in Data Studio
No dataset card yet
- Downloads last month
- 11