langgraph_mcp

langgraph_mcp

0
developer_tools

LangGraph Documentation MCP Server is an MCP server providing access to LangGraph documentation with vector store-based retrieval using Ollama embeddings for document vectorization. It supports both semantic search tools and full documentation access.

LangGraph Documentation MCP Server

This project implements a Mod Control Protocol (MCP) server that provides access to LangGraph documentation through a vector store-based retrieval system. The implementation is based on the MCP From Scratch tutorial and has been updated to use Ollama for embeddings.

Features

  • Vector store-based document retrieval using SKLearnVectorStore
  • Ollama embeddings for document vectorization
  • MCP server implementation with FastMCP
  • Document loading and processing from LangGraph documentation
  • Support for both tool-based queries and full documentation access

Prerequisites

  • Python 3.12+
  • Ollama installed and running locally (default port: 11434)
  • Required Python packages
    • langchain_community langchain-anthropic langchain_ollama scikit-learn bs4 pandas pyarrow matplotlib lxml langgraph tiktoken "mcp[cli]

Installation

  1. Clone the repository
  2. Create and activate a virtual environment:
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
  1. Install dependencies:
pip install langchain_community langchain-anthropic langchain_ollama scikit-learn bs4 pandas pyarrow matplotlib lxml langgraph tiktoken "mcp[cli]"

Project Structure

  • langgraph_mcp.py: Main MCP server implementation
  • build-tool.ipynb: Jupyter notebook for building and testing the vector store
  • llms_full.txt: Generated documentation file
  • sklearn_vectorstore.parquet: Vector store file

Usage

  1. First, build the vector store using the Jupyter notebook:
jupyter notebook build-tool.ipynb
  1. Run the MCP server:
python langgraph_mcp.py

Available Tools

langgraph_query_tool

A tool that queries the LangGraph documentation using semantic search:

@mcp.tool()
def langgraph_query_tool(query: str):
    """
    Query the LangGraph documentation using a retriever.
    
    Args:
        query (str): The query to search the documentation with

    Returns:
        str: A str of the retrieved documents
    """

Full Documentation Access

Access the complete LangGraph documentation through the resource endpoint:

@mcp.resource("docs://langgraph/full")
def get_all_langgraph_docs() -> str:
    """
    Get all the LangGraph documentation.
    """

Implementation Details

The project uses:

  • Ollama embeddings with the nomic-embed-text model
  • SKLearnVectorStore for document storage and retrieval
  • BeautifulSoup for HTML parsing
  • RecursiveUrlLoader for documentation scraping
  • RecursiveCharacterTextSplitter for document chunking

Credits

This implementation is based on:

License

This project is open source and available under the MIT License.