MCP Python SDK 추가하기

Your project is scaffolded, but it can't speak MCP yet. The next step is pulling in the official library that does all the heavy lifting. 📦

The official Python SDK ships as a package simply named mcp. It contains everything you need to build servers and clients in plain Python.

Inside your project folder, run uv add to install the SDK and record it as a dependency. One command grabs it and updates your project files.

The [cli] extra adds the command-line tools that run and inspect servers. Without it you get the library but miss the handy mcp commands.

Always wrap the name in quotes, like "mcp[cli]". Many shells treat square brackets as special characters, and quotes keep them literal.

If you prefer plain pip, the same package installs the same way. Either tool gives you an identical SDK, so pick whichever fits your workflow.

uv add doesn't just install; it writes mcp into your pyproject.toml. Anyone who clones your project gets the exact same requirement.

uv also updates a uv.lock file that pins the exact resolved versions. That makes your installs reproducible on any machine, every time.

Verify the install worked by importing the SDK inside your environment. If nothing errors out, the mcp package is ready to use.

The class you'll use most is FastMCP, found under mcp.server.fastmcp. It's the high-level way to define tools without touching raw protocol details.

That single import unlocks decorators for tools, resources, and prompts. The SDK turns ordinary Python functions into capabilities an AI can call. ✨

You want the SDK plus its command-line helpers.

You added mcp[cli], recorded it in pyproject.toml, locked the versions, and confirmed the import. Your project can now build a real MCP server. 🎯