McpDocServer

What is McpDocServer?

McpDocServer is an MCP (Model Context Protocol) server for development framework documentation. It provides multi-threaded documentation crawling, local document loading, keyword search, and document detail retrieval. It is designed to solve AI hallucination problems in Cursor by delivering accurate, up-to-date API information directly from official sources.

How to use McpDocServer?

Clone the repository, optionally set PUPPETEER_SKIP_DOWNLOAD to skip Puppeteer’s Chrome download, then run npm install. Create a crawler config in config/doc-sources.js and run npm run crawl to fetch documentation into the docs/ folder. Start the server with npm start. The server exposes two MCP tools: search_docs (search with optional source/limit) and get_doc_detail (get page content by ID). Integrate with Cursor by adding the absolute path to server.js in your mcp.json.

Key features of McpDocServer

• Load documentation from local JSON files
• Full-text keyword search across all loaded sources
• Retrieve detailed content of a specific document page
• Crawl external documentation websites and convert to local format
• Reload all documents by searching for “reload”
• Multi-threaded crawler with configurable concurrency and retries

Use cases of McpDocServer

• Provide Cursor AI with precise, real-time framework API documentation
• Eliminate outdated or nonexistent API recommendations in code generation
• Reduce version confusion by referencing the exact documentation source
• Offer accurate method signatures and parameter lists for a framework
• Integrate official documentation into any MCP-compatible agent or IDE

FAQ from McpDocServer

What frameworks can I use with McpDocServer?

Any framework or library is supported — you only need to create a crawler configuration with the site URL, include/exclude patterns, and run the crawler.

Where does the documentation data live?

All crawled documentation is stored as JSON files in the docs/ directory. The server automatically loads every JSON file in that folder on startup.

How do I reload documentation without restarting the server?

Call the search_docs tool with query set to "reload". This will trigger a full reload of all document sources.

What are the runtime dependencies?

Node.js, plus the puppeteer package for crawling. Optionally set PUPPETEER_SKIP_DOWNLOAD if you already have Chrome installed.

What MCP tools does the server provide?

Two tools: search_docs (required parameter query, optional source and limit) and get_doc_detail (required id, optional source).