mcpdoc

Install

{
  "mcpServers": {
    "langgraph-docs-mcp": {
      "command": "uvx",
      "args": [
        "--from",
        "mcpdoc",
        "mcpdoc",
        "--urls",
        "LangGraph:https://langchain-ai.github.io/langgraph/llms.txt LangChain:https://python.langchain.com/llms.txt",
        "--transport",
        "stdio"
      ]
    }
  }
}

This is third-party code. Review the repository files before installing.

What it does

The server takes a list of llms.txt files and gives the agent two tools: listing documentation sources and fetching pages linked from those files. You can see exactly which pages the agent read and what it got, unlike a client's built-in web search. Access is restricted by domain: a remote llms.txt allows only its own domain, and a local file needs domains listed explicitly. It runs over stdio or SSE and starts via uvx without installation.

Who it is for. For developers who want the agent to take library docs only from chosen sources.

Example request

What types of memory does LangGraph have? Check the docs through llms.txt

Limitations

Works only with sources that publish llms.txt. Requires uv. Claude Desktop may hit Python version issues, in which case you pass the interpreter path explicitly. The global rules the authors suggest are not supported by every client. The server is local; access from Russia depends only on the documentation sites.

How to disable. Remove the server block, for example langgraph-docs-mcp, from the MCP config and delete related rules from client settings.

MCP

Transport

stdio, sse

Authentication

not required

Security check

• Fetches web pages linked from llms.txt

README in short

The README explains why a dedicated llms.txt server helps and links LangGraph and LangChain files. It then covers installing uv, domain access rules, local testing with MCP Inspector, and connecting Cursor, Windsurf, Claude Desktop and Claude Code with sample agent rules. MIT licensed.

Related