How to Intermediate · 3 min read

How to use MCP with Claude API

Q: How to use MCP with Claude API

Use the mcp Python SDK to create an MCP server that connects to the Claude API by running a stdio_server with your Anthropic client. This enables AI agents to interact with external tools via the MCP protocol seamlessly.

Quick answer

Use the mcp Python SDK to create an MCP server that connects to the Claude API by running a stdio_server with your Anthropic client. This enables AI agents to interact with external tools via the MCP protocol seamlessly.

PREREQUISITES

Python 3.8+
Anthropic API key
pip install anthropic>=0.20 mcp>=0.1

Setup

Install the required packages and set your environment variable for the Anthropic API key.

Install packages: pip install anthropic mcp
Set environment variable: export ANTHROPIC_API_KEY='your_api_key' on Linux/macOS or setx ANTHROPIC_API_KEY "your_api_key" on Windows.

bash

pip install anthropic mcp

Step by step

This example shows how to create an MCP stdio server that uses the Claude API to handle AI agent requests. The server reads from standard input/output and connects to the claude-3-5-sonnet-20241022 model.

python

import os
from anthropic import Anthropic
from mcp.server.stdio import stdio_server

client = Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])

# Define a handler function that sends messages to Claude via Anthropic SDK
async def handler(messages, system):
    response = await client.messages.create(
        model="claude-3-5-sonnet-20241022",
        max_tokens=1024,
        system=system,
        messages=messages
    )
    return response.content[0].text

# Run the MCP stdio server with the handler
if __name__ == "__main__":
    stdio_server(handler)

Common variations

You can customize the MCP integration by:

Using different Claude models like claude-sonnet-4-5 for improved coding or reasoning.
Running the MCP server asynchronously in an event loop if integrating into larger async applications.
Connecting MCP to other transport layers besides stdio, such as SSE or WebSocket, depending on your deployment.

python

import asyncio

async def async_handler(messages, system):
    response = await client.messages.create(
        model="claude-sonnet-4-5",
        max_tokens=1024,
        system=system,
        messages=messages
    )
    return response.content[0].text

async def main():
    await stdio_server(async_handler)

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

Troubleshooting

If you see connection errors or timeouts, verify your ANTHROPIC_API_KEY is set correctly and has permissions. For JSON decode errors, ensure the MCP messages conform to the protocol. Use logging to debug message flow between MCP and Claude.

✅

Key Takeaways

Use the official mcp Python SDK to run an MCP server with Claude API integration.
The stdio_server function enables easy local testing and deployment of MCP agents.
Customize the handler to use different Claude models or async patterns as needed.
Always secure your ANTHROPIC_API_KEY in environment variables for production.
Check MCP protocol compliance and logs to troubleshoot communication issues.

Verified 2026-04 · claude-3-5-sonnet-20241022, claude-sonnet-4-5

Verify ↗