Building a Python MCP Server

Python is the most popular language for MCP servers. The modern way is FastMCP, a decorator-based wrapper over the official SDK.

Setup

mkdir calc-py && cd calc-py
python3 -m venv .venv && source .venv/bin/activate
pip install mcp

The server

Create server.py:

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("calc")

@mcp.tool()
def multiply(a: float, b: float) -> str:
    """Multiply two numbers and return the product."""
    return str(a * b)

@mcp.tool()
def add(a: float, b: float) -> str:
    """Add two numbers."""
    return str(a + b)

if __name__ == "__main__":
    mcp.run()  # defaults to stdio transport

Run it: python server.py.

Register it

{
  "mcpServers": {
    "calc-py": {
      "command": "python",
      "args": ["/full/path/to/server.py"]
    }
  }
}

FastMCP niceties

• Docstrings become tool descriptions — the model reads them.
• Type hints become the JSON schema — no manual schema writing.
• You can also expose resources:

@mcp.resource("file://notes")
def notes() -> str:
    return "Buy milk. Ship the feature."

Python vs TypeScript

• Python (FastMCP): fastest to write, best for data/ML and quick scripts.
• TypeScript SDK: first-class, great for Node projects and the richest type story.
• Both speak the same protocol — a client can't tell the difference.

The model calls multiply by name, reads your docstring to understand it, and validates arguments against the type hints. Docstrings are not optional decoration — they're the UI.