MCP Doc Server

Clone McpDocServer and install Node.js dependencies. Set PUPPETEER_SKIP_DOWNLOAD=true so it uses your system Chrome instead of downloading a separate Chromium binary.

git clone https://github.com/ruan11223344/McpDocServer.git
cd McpDocServer
export PUPPETEER_SKIP_DOWNLOAD=true
npm install

Edit config/doc-sources.js to define which documentation sites to crawl. Each source needs a name, base URL, and optional include/exclude URL patterns. The name you assign becomes the 'source' parameter in search queries.

// config/doc-sources.js
export const docSources = [
  {
    name: "react",
    url: "https://react.dev/reference",
    includePatterns: ["/reference/"],
    excludePatterns: [/blog/, /learn/]
  }
];

export const crawlerConfig = {
  maxConcurrency: 20,
  pageLoadTimeout: 30000,
  contentLoadTimeout: 5000,
  headless: true,
  maxRetries: 3,
  retryDelay: 2000,
  requestDelay: 1000
};

Run the crawler to fetch and index the configured documentation sites. This may take several minutes for large sites. The indexed documents are stored locally.

npm run crawl

Start the MCP server process. It serves the search and retrieval tools over the MCP protocol.

npm start

Add McpDocServer to your Claude Desktop or Cursor config. The server is launched as a Node.js process pointing to the built server.js file.

{
  "mcpServers": {
    "doc-server": {
      "command": "node",
      "args": ["/absolute/path/to/McpDocServer/server.js"],
      "env": {
        "NODE_ENV": "development"
      }
    }
  }
}