mcp-python

MCP Servers in Python

Create and extend MCP servers in Python with FastMCP and the official MCP Python SDK. Based on MCP Build a server and MCP Best Practices.

When to Use

• Building or extending MCP servers in Python
• Adding tools, resources, or prompts
• Debugging MCP server connection or tool discovery
• Choosing transport (stdio vs streamable HTTP)
• Configuring clients (Claude Desktop, Cursor) to run the server

Core Concepts

• Tools: Functions callable by the LLM (with user approval). Have explicit input/output and side-effect disclosure.
• Resources: Readable data (files, API responses) the client fetches for context. URI-based.
• Prompts: Pre-written templates for specific tasks; reduce prompt drift.
• Discovery: Clients enumerate tools/resources/prompts and get schemas at connect time.
• Transports: stdio for local, per-user processes; streamable HTTP for remote, shared services.

FastMCP Quick Reference

Python 3.12, MCP Python SDK 1.2.0+. Use uv add "mcp[cli]" or uv add fastmcp (or project may use fastmcp package).

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("my-server")


@mcp.tool()
async def get_forecast(latitude: float, longitude: float) -> str:
    """Get weather forecast for a location.
    Args:
        latitude: Latitude of the location
        longitude: Longitude of the location
    """
    # ... fetch and return string
    return result


@mcp.resource("config://{key}")
def get_config(key: str) -> str:
    """Get config value by key."""
    return config_store.get(key, "")


@mcp.prompt()
def plan_task(goal: str, steps: int = 5) -> str:
    """Generate a step-by-step plan. Args: goal, steps (default 5)."""
    return f"Plan for: {goal} in {steps} steps."


def main():
    mcp.run(transport="stdio")


if __name__ == "__main__":
    main()

• Tool schemas: FastMCP infers names, types, and descriptions from type hints and docstrings (Args/Returns).
• Run: mcp.run(transport="stdio") for stdio; use streamable HTTP for remote deployment.