Skip to content

16.05 Connecting the Math and Weather Servers

Before we write the client itself, make sure both of the servers we built are currently running. Open two separate terminal windows and run:

(Terminal 1) Math Server: 

uv run servers/math_server.py

(Terminal 2) Weather Server: 

uv run servers/weather_server.py
(You should see the weather server binding to local port 8000).

The Client Setup

Create a new file named langchain_client.py in the root of your project.

We are going to use a special abstraction provided by the LangChain team: The MultiServerMCPClient.

By default in the MCP specification, the connection between a client and a server is 1-to-1. If you wanted an agent to access 5 different MCP servers, you would traditionally have to write 5 distinct client connection classes and manage a massive mess of async context managers to bind them all together.

LangChain abstracts this completely.

The Boilerplate

import asyncio
import os
from dotenv import load_dotenv

# LangChain Imports
from langchain_mcp_adapters.client import MultiServerMCPClient
from langgraph.prebuilt import create_react_agent
from langchain_openai import ChatOpenAI

load_dotenv()

async def main():
    llm = ChatOpenAI(model="gpt-4o")
    print("Agent Initialized.")

    # We will instantiate the Multi-Client here...

if __name__ == "__main__":
    asyncio.run(main())

Run uv run langchain_client.py to ensure your boilerplate executes cleanly without throwing any dependency errors. In the next lesson, we will initialize the Multi-Client.