FastMCP

FastMCP is the standard Python framework for the Model Context Protocol (MCP), the open protocol that connects LLMs to external tools and data. It turns a plain Python function into an MCP-compliant tool with a single decorator, auto-generating schema, validation, and documentation, and provides matching abstractions for MCP clients and interactive in-conversation apps.

Originally created by Jeremiah Lowin, FastMCP 1.0 was incorporated into the official MCP Python SDK in 2024; the actively maintained standalone project (now backed by Prefect) has since grown into the dominant MCP framework, reportedly downloaded a million times a day and powering roughly 70% of MCP servers across all languages, with a companion commercial gateway (Prefect Horizon) for production deployment.

What You Get

• @mcp.tool / @mcp.resource / @mcp.prompt decorators that turn plain Python functions into MCP-compliant capabilities with auto-generated schemas
• A full-featured MCP client for connecting to any server (local subprocess, remote HTTP/SSE) programmatically or via CLI
• Built-in auth provider integrations and transport handling so servers don’t hand-roll protocol plumbing
• An Apps layer for rendering interactive UI directly inside an LLM conversation, beyond plain text tool results
• A CLI (fastmcp run, fastmcp install, fastmcp dev) for running, installing, and debugging MCP servers

Common Use Cases

• Exposing internal APIs, databases, or scripts as MCP tools for Claude, ChatGPT, or other MCP-compatible LLM clients
• Building MCP servers that need production concerns like auth, structured error handling, and multiple transports (stdio, HTTP, SSE)
• Writing MCP clients to programmatically call tools on existing MCP servers as part of an agent pipeline
• Prototyping agent tools quickly, since a decorated Python function is enough to get a working MCP server

Under The Hood

Architecture - The package (now organized under fastmcp_slim/fastmcp, ~75k lines of Python) is split into clear subsystems: server/ for the core FastMCP server object and tool/resource/prompt registration, client/ for the MCP client implementation, cli/ for the fastmcp command (run.py, install/, discovery.py, generate.py, auth.py), apps/ for the interactive-UI layer, contrib/ for community extensions, and experimental/ for in-progress features; mcp_config.py centralizes server configuration and telemetry.py handles opt-in usage metrics. Tech Stack - Built on the official low-level MCP Python SDK, targeting Python 3.10+, packaged with hatchling and uv-dynamic-versioning, dependency-managed via uv.lock; the repo also ships a separate fastmcp_remote package for remote-server tooling. Code Quality - Test coverage is extensive and organized by subsystem (tests/server, tests/client, tests/tools, tests/resources, tests/prompts, tests/cli, tests/conformance, tests/integration_tests, tests/telemetry), with a dedicated tests/deprecated suite tracking backward-compatibility guarantees across the 1.0→3.x evolution — a strong signal of discipline for a fast-moving project with 100+ releases. API Design - The signature design goal is near-zero boilerplate: from fastmcp import FastMCP; mcp = FastMCP(...) plus a @mcp.tool decorator is enough to produce a working server, with schema/validation/docs generated automatically from type hints and docstrings — explicitly built to make MCP best practices the path of least resistance rather than something developers must assemble themselves.