mcpdoc

mcpdoc is an MCP server that enables AI assistants to access website documentation formatted according to the llms.txt standard over the Model Context Protocol. It connects AI coding agents such as Claude Desktop, Cursor, and Windsurf directly to external documentation portals, utilizing structured llms.txt files that web platforms provide specifically for language models. Developers and technical researchers use mcpdoc to retrieve up-to-date API references, libraries, and framework manuals directly during development workflows, eliminating outdated training cutoff constraints. Similar to how robots.txt directs search crawler behavior, llms.txt files present a curated breakdown of target URLs and concise summaries tailored for artificial intelligence consumption. By pointing mcpdoc to specified llms.txt endpoints—such as documentation for LangGraph or LangChain—the server parses and serves the underlying content to the calling client. This setup provides targeted knowledge retrieval and prevents hallucinations when generating code against rapidly changing open-source libraries, ensuring that agents query authoritative upstream documentation files rather than broad or unverified web scrapes.

Category: Developer Tools & Code Intelligence

Tags: documentation, knowledge-retrieval, llmstxt, web scraping

Visit mcpdoc

How to install and configure mcpdoc

1. Initialize a project directory and virtual environment using uv by running uv init, uv venv, and source .venv/Scripts/activate. 2. Install the necessary project dependencies by running uv sync. 3. Determine the path to your uv binary using which uv and the absolute path to your local project clone using pwd. 4. Open your Claude Desktop configuration file and register the server under mcpServers: json { "mcpServers": { "langgraph-docs-mcp": { "command": "/path/to/uvx", "args": [ "--from", "/path/to/mcpdoc", "mcpdoc", "--urls", "LangGraph: https://langchain-ai.github.io/langgraph/llms.txt LangChain: https://python.langchain.com/llms.txt", "--transport", "stdio", "--port", "8081", "--host", "localhost" ] } } } 5. Replace the file paths with the output of which uv and pwd, then restart Claude Desktop.

What you can do with mcpdoc

• Fetching current LangChain and LangGraph API reference docs directly inside Claude Desktop via indexed llms.txt feeds. - Providing coding agents in Cursor or Windsurf with structured documentation breakdowns to reduce hallucinated function calls. - Monitoring and inspecting library updates across multiple software frameworks from a single MCP endpoint configuration. - Parsing curated documentation URLs from any site publishing an llms.txt file to guide agentic code generation.

Key facts

• https://github.com/tevinric/mcpdoc
• Developer Tools & Code Intelligence, Web Search & Research
• documentation, knowledge-retrieval, llmstxt, web scraping

Part of MCP Servers

What is mcpdoc?

mcpdoc is a Model Context Protocol server designed to fetch and serve website documentation structured in the llms.txt format. It allows artificial intelligence agents like Claude Desktop, Cursor, and Windsurf to consume the latest web page breakdowns and API guides directly from upstream documentation publishers.

What can mcpdoc do?

mcpdoc parses llms.txt files from configured web domains, extracts structured lists of URLs and their descriptions, and exposes this content over standard Model Context Protocol transports. This enables AI tools to retrieve accurate, current documentation for software frameworks such as LangChain and LangGraph during development sessions.

Which MCP clients work with mcpdoc?

mcpdoc works with MCP-compliant desktop and editor clients including Claude Desktop, Windsurf, and Cursor. Because it supports stdio transport, it can be added to any standard MCP client configuration file that executes local commands using the uv package manager.

How do I configure documentation URLs in mcpdoc?

Documentation endpoints are configured using the --urls argument in your client configuration. You pass library names paired with their respective llms.txt links, such as passing LangGraph and LangChain URLs separated by spaces. The server reads these locations upon launch to make documentation accessible to the assistant.