MCP Crash Course

MCP Crash Course

A demonstration project showcasing the integration of LangChain with Model Control Protocol (MCP) adapters. This project implements a system that can handle both mathematical calculations and weather queries through separate MCP servers.

Features

• Multiple MCP server integration (math and weather servers)
• LangChain integration with OpenAI
• Async operation support
• Environment variable configuration

Prerequisites

• Python 3.x
• OpenAI API key

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
   

3. Install dependencies:

   pip install -r requirements.txt
   

4. Create a .env file and add your OpenAI API key:

   OPENAI_API_KEY=your_api_key_here
   

Project Structure

• main.py - Single server implementation using stdio client
• langchain_client.py - Multi-server implementation using MCP client
• server/ - Directory containing MCP server implementations
  • math_server.py - Server for mathematical operations
  • weather_server.py - Server for weather queries

Usage

1. Run the single server example:

   python main.py
   

2. Run the multi-server example:

   python langchain_client.py
   

How It Works

The project demonstrates two approaches to using MCP:

1. Single Server (main.py):

   • Uses stdio client for communication
   • Connects to a math server for calculations
   • Implements a React agent using LangChain

2. Multi-Server (langchain_client.py):

   • Uses MultiServerMCPClient for managing multiple servers
   • Connects to both math and weather servers
   • Allows the agent to choose appropriate tools based on the query

Dependencies

• langchain-core
• langchain-openai
• langchain-mcp-adapters
• langgraph
• python-dotenv
• mcp

License

MIT