MCP Python SDK — MCP protocol

Official SDK for building MCP servers and clients in Python

Basic Information

What is this program?

MCP Python SDK is the official SDK for building MCP servers and clients in Python. Using the decorator API, you can complete a server that exposes a single tool in about 20 lines.

It follows familiar Python patterns like FastAPI and Flask, and supports stdio and Streamable HTTP transports with AnyIO-based asynchronous I/O. It is a natural choice when exposing data analysis and machine learning tools via MCP.

What problems does it solve?

• Situations where you need to expose internal tools/models built in Python via MCP
• Situations where you need to integrate MCP servers with FastAPI and data pipelines
• Situations where you need to distribute using Python ecosystem tools such as uv and pip

Key Features

Decorator API

Define tools easily with the @mcp.tool() decorator

• Use case: Build a server with minimal code
• Result: A Python function becomes an MCP tool
• Limitations: Requires understanding asynchronous function patterns

Multiple transports

Supports stdio and Streamable HTTP transports

• Use case: Implement both local and remote servers
• Result: Can integrate with FastAPI
• Limitations: Authentication setup required for remote deployment

Built-in CLI

Run and test servers with the mcp CLI

• Use case: Fast development loop
• Result: Easy distribution via uvx and pipx
• Limitations: CLI options vary by version

AnyIO-based

Supports Python async runtimes such as asyncio and trio

• Use case: Integrate with existing asynchronous code
• Result: Event loop compatibility
• Limitations: Synchronous code requires separate handling

Installation and Execution

uv (Recommended)

Install quickly with the uv package manager.

uv add "mcp[cli]"

pip

Install with pip.

pip install "mcp[cli]"

pipx

Install as an isolated CLI tool.

pipx install "mcp[cli]"

Real Usage Examples

Creating a Server

Define a tool with a decorator and run the server.

@mcp.tool() def add(a:int,b:int)->int: return a+b

Running the Server

Start the server with stdio transport.

mcp run server.py

Client Connection

Connect to remote/local servers with McpClient.

async with client_session(...) as client: await client.list_tools()

Advantages and Limitations

Advantages

• MIT License, the largest Python MCP ecosystem
• Build a minimal-code server with decorators
• Natural integration with FastAPI and AnyIO
• Compatible with Python distribution tools such as uv and pipx

Considerations Before Use

• Limited for large-scale concurrency due to the Python GIL
• Some advanced features may lag compared to the TypeScript SDK
• Requires learning asynchronous patterns

Comparison with Similar Programs

Official Resources and Further Learning

• Official website — https://modelcontextprotocol.io/docs/sdk
• Official documentation / knowledge base — https://github.com/modelcontextprotocol/python-sdk
• Installation/Distribution — uv add "mcp[cli]" or pip install "mcp[cli]"
• Release notes — https://github.com/modelcontextprotocol/python-sdk/releases
• Source / development materials — https://github.com/modelcontextprotocol/python-sdk

Version and supported environment information is based on the official site as of 2026-10-18. Installation commands may vary depending on the distribution and version you use, so also check the official installation documentation.

Actual Execution Screen

Synthetic example — not an actual execution result. The screen is a synthetic example and is not an actual execution result.

!MCP Python SDK installation and server execution example

Server actually in operation

Source (HTML): https://games.aidebatehub.com/programs/mcp/protocol/py-sdk.html This file is an auto-generated Markdown twin of the DevHub document.