MCP + LangGraph Agent

This is a minimal, functional example of an agent powered by LangGraph with tools implemented using MCP (Model Context Protocol) servers instead of traditional LangChain tools. It's based on the official LangGraph tutorial, adapted to demonstrate how to integrate MCP servers as tools.

🧠 Use this repo as a skeleton to quickly build your own LangGraph agent with MCP tools!

Features

flowchart TD
    __start__ --> chatbot
    chatbot -.-> tools
    chatbot -.-> __end__
    tools --> chatbot

• Integrates LangGraph for managing agent state and message routing.
• Uses MCP servers to provide access to tools.
• Includes example MCP servers for math and weather.
• Provides a command-line interface for interacting with the agent.

Installation

1. Clone the repository:

   git clone <repository_url>
   

2. Install the dependencies using Poetry:

   poetry install
   

Usage

1. Set the Anthropic API key in the .env file. You may need to create this file if it doesn't exist. For example:

   ANTHROPIC_API_KEY=your_api_key
   

2. Start the MCP servers:

   • Math server: python src/mcp_servers/math_server.py
   • Weather server: python src/mcp_servers/weather_server.py

3. Run the agent:

   poetry run main
   

   This will start the agent in interactive mode. You can then enter prompts, and the agent will respond using the tools provided by the MCP servers.

Example

User: What's (3 + 5) x 12?
Assistant: The result of (3 + 5) × 12 = 96
User: What is the weather in New York?
Assistant: It's always sunny in New York.

MCP Server Configuration

The MCP servers are configured in src/main.py. You can modify the configuration to add or remove servers, or to change the transport mechanism.

LLM Configuration

The LLM used by the agent can be changed in src/common.py.