How MCP client server communication works
Quick answer
The
MCP client-server communication uses a protocol where the Server manages requests and responses over standard input/output or SSE transport. The mcp Python SDK provides classes like Server and stdio_server to implement this communication, enabling AI agents to interact with external tools or resources seamlessly.PREREQUISITES
Python 3.8+pip install mcpBasic knowledge of Python async programming
Setup
Install the official mcp Python SDK and prepare your environment to run MCP client-server communication.
pip install mcpStep by step
This example demonstrates a simple MCP server using the stdio_server transport that listens for client messages and responds accordingly.
import asyncio
from mcp.server import Server
from mcp.server.stdio import stdio_server
async def handle_request(request):
# Process the incoming request and return a response
return {"result": f"Received: {request.get('input', '')}"}
async def main():
server = Server(handle_request)
await stdio_server(server)
if __name__ == "__main__":
asyncio.run(main())Common variations
You can implement MCP communication over different transports such as SSE or TCP by using other server implementations. Async handling allows concurrent requests. You can also customize the handle_request function to integrate with AI models or external APIs.
import asyncio
from mcp.server import Server
from mcp.server.stdio import stdio_server
async def handle_request(request):
# Example: echo input with uppercase
user_input = request.get('input', '')
return {"result": user_input.upper()}
async def main():
server = Server(handle_request)
# Replace stdio_server with your custom transport if needed
await stdio_server(server)
if __name__ == "__main__":
asyncio.run(main())Troubleshooting
- If the server does not respond, ensure your client sends properly formatted JSON messages over the expected transport.
- Check that
asyncio.run()is used correctly to start the server. - For debugging, add logging inside
handle_requestto trace incoming requests.
Key Takeaways
- Use the official
mcpPython SDK to implement client-server communication for AI agents. - The
Serverclass handles requests asynchronously over stdio or other transports. - Customize the request handler to integrate AI models or external tools.
- Always run the server with
asyncio.run()for proper async execution. - Debug communication by logging requests and verifying JSON message formats.