Getting Started with FastMCP in Python

FastMCP is the high-level Python framework for building MCP servers (and clients)—ordinary functions become tools with schema generation handled for you. Official docs: gofastmcp.com (not an old GitHub README as your only source). Repo reference: github.com/PrefectHQ/fastmcp.

After work, FastMCP is the fastest path from “I understand MCP” to “I shipped one tool.”

Who this is for

Python learners who already skimmed MCP concepts and want to write a server tonight. If you have never written a typed function, do Intro first—decorators will feel less magical.

Overview: Top 10 AI agent tools for after-work Python.

Steps to start tonight

1. Install from current docs

Prefer the install line on gofastmcp.com/getting-started/installation:

# recommended in current docs:
uv add fastmcp
# or:
pip install -U fastmcp
fastmcp version

2. Create my_server.py with one tool

From the quickstart:

from fastmcp import FastMCP

mcp = FastMCP("AfterWork Demo")

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

if __name__ == "__main__":
    mcp.run()

Adjust decorator/API details if the live quickstart differs—docs win.

3. Run locally (stdio)

python my_server.py
# or: fastmcp run my_server.py:mcp

4. Call the tool once

Use MCP Inspector, a FastMCP client snippet from the docs, or your preferred host. Confirm add(2, 3) returns 5.

5. Stop before cloud deploy

Do not deploy to Horizon/cloud or add OAuth tonight. Local green run is the win.

Tiny exercise (20–30 minutes)

Add greet(name: str) -> str. Call both tools. If schemas look wrong, fix type hints and docstrings—do not paste a giant prompt into the tool body.

Pitfalls

• Using an archived blog / old monorepo as the sole source. Start at gofastmcp.com.
• Confusing FastMCP with the low-level MCP SDK APIs. FastMCP is the ergonomic layer.
• Shipping HTTP + auth before stdio works.
• Hard-coding API keys inside tools.
• Ten tools and a RAG stack on night one. One add is enough.

Why this fits after work

You practice MCP as normal Python: functions, types, one process. That skill compounds in every agent host you use later.

Next step

When tools feel ordinary, deepen Python with Intro to Python Programming so larger FastMCP servers stay readable.