FastMCP
A Python MCP application framework for building servers, clients, and interactive apps through one API, with stdio and Streamable HTTP transports.
Project overview
FastMCP is a Python application framework for the Model Context Protocol. It puts server-side tool, resource, and prompt declarations, client connections, and interactive apps behind one Python API. A minimal server creates a FastMCP instance and registers a typed function; stdio works locally, while Streamable HTTP supports remote deployments.
Core capabilities
Define MCP tools, resources, and prompts with Python functions
Create a FastMCP instance and use decorators to register typed Python functions as tools while organizing resources and prompts in the same framework.
Build MCP clients for local or remote servers
The Client API supports in-process connections, stdio subprocesses, and HTTP endpoints for testing or orchestration.
Extend servers with middleware, authentication, providers, and transforms
Server documentation describes extension points for controlling requests, capability sources, and exposed behavior.
Build interactive MCP apps for supported clients
FastMCP Apps can provide in-conversation interfaces inside clients that support the capability.
Setup and connection
Install FastMCP with uv or pip in an isolated Python 3.10 or newer environment, then create a minimal server and launch it over stdio.
Let an AI Agent install it
Send this prompt to Codex, Claude Code, or another AI agent that can work with your local environment.
Help me install FastMCP from https://github.com/PrefectHQ/fastmcp. First read README.md, docs/getting-started/installation.mdx, docs/getting-started/quickstart.mdx, and pyproject.toml. Confirm that Python is at least 3.10 and use an isolated environment. Follow the documented uv add fastmcp command; if the project does not use uv, use pip install fastmcp instead. Explain and recommend an exact compatible pin before changing a production project's lockfile, and do not overwrite existing dependency files without permission. After installation, run fastmcp version, create the official minimal greet server, and run fastmcp run my_server.py:mcp. Report the installed version, changed files, startup result, and any unverified items. Ask before requesting credentials, expanding file or network access, opening an HTTP port, overwriting files, or taking any risky action.- Python 3.10 or newer
- uv or pip and an isolated Python virtual environment
- Permission to create a Python file and start a local process
uv add fastmcp- 1Install and check the CLI
Run uv add fastmcp in an isolated environment, then run fastmcp version. If using pip, follow the documented alternative command pip install fastmcp.
- 2Create a minimal MCP server
Create my_server.py, import FastMCP, set mcp = FastMCP("My MCP Server"), and decorate a typed Python function that accepts a name and returns a greeting with @mcp.tool.
- 3Launch over stdio
Run fastmcp run my_server.py:mcp. stdio is the default transport and works for a local MCP client that starts the server as a subprocess.
- 4Switch to HTTP when needed
For a remote or self-hosted deployment, run fastmcp run my_server.py:mcp --transport http --port 8000. Configure authentication, error masking, and network controls before exposure.
Run fastmcp version to confirm the CLI is available. Save the quickstart example as my_server.py and run fastmcp run my_server.py:mcp. A process that remains running without import or startup errors passes the minimal installation check.
- Pin an exact tested version in production; v4.0.3 was the latest stable release when this page was verified.
- Prefer stdio or Streamable HTTP for new projects; SSE is a legacy transport.
- Install the tasks extra only when background task support is required.
Use cases
Wrap internal Python capabilities as MCP tools
Expose typed inputs and outputs from existing functions to agents while keeping business dependencies and access controls inside the server process.
Write automated clients for MCP services
Use the Client API in tests or orchestration code to list and call tools through in-process, stdio, or HTTP connections.
Deploy an authenticated remote MCP endpoint
Combine Streamable HTTP, middleware, and authentication extensions for a self-hosted service while retaining responsibility for network and credential boundaries.
Assessment
FastMCP's main advantage is one Python API spanning server, client, and app development, with a clear path from local stdio to HTTP deployment. It is still a framework that executes business code with process-level permissions, so production adoption requires version pins, transport and authentication review, and an audit of each real tool.
Why it may be useful
- The minimal example is short and registers ordinary Python functions as MCP tools.
- It covers servers, clients, apps, middleware, authentication, and providers in one project.
- Installation, quickstart, and core API documentation are available, and the repository remains actively maintained.
What to know first
- It does not provide managed hosting, default business-level isolation, or automatic security review.
- Production deployments must handle authentication, network exposure, credentials, and tool permissions.
- The docs warn that ecosystem compatibility can cause minor-version breaks, while the current SECURITY.md support table still trails the v4 release.
README
FastMCP
Overview
A Python MCP application framework for building servers, clients, and interactive apps through one API, with stdio and Streamable HTTP transports. 🚀 The fast, Pythonic way to build MCP servers and clients.
Getting started
- Install FastMCP with uv or pip in an isolated Python 3.10 or newer environment, then create a minimal server and launch it over stdio.
- Servers can expose Python functions, resources, and prompt templates as MCP capabilities, with authentication, middleware, providers, and transforms available as extensions.
- Clients can connect to in-memory servers, stdio subprocesses, or remote HTTP endpoints and use an asynchronous API to list and call capabilities.
- Run fastmcp version to confirm the CLI is available. Save the quickstart example as my_server.py and run fastmcp run my_server.py:mcp. A process that remains running without import or startup errors passes the minimal installation check.
Configuration
uv add fastmcp