Python MCP Workshop

This workshop shows how to run a small Python MCP server and connect it to GitHub Copilot in VS Code.

The server provides two tools:

hello_world() returns Hello, world!.
add_numbers(first_number: float, second_number: float) adds two numbers together.

You do not need deep MCP knowledge for this workshop. The goal is to get the server running, connect it to Copilot, and try the tools from chat.

Tool implementations live in separate files under src/mcp_workshop/tools/, with their MCP title, description, and hints defined next to the function. The central inventory in src/mcp_workshop/tools/inventory.py collects those tools and registers them with the server.

Local setup

You need Python 3.10 or newer. If you are using Codespaces, the devcontainer already provides a suitable Python version.

From the repository root, create a virtual environment and install the project:

python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
pip install -e .[dev]

On Windows PowerShell, activate the environment with:

.venv\Scripts\Activate.ps1

Run the server

This workshop server supports two MCP transports:

stdio: best when GitHub Copilot or MCP Inspector starts the server for you.
HTTP: useful when you want a local URL such as http://127.0.0.1:8000/.

Option 1: stdio mode

Stdio mode is the default:

python -m mcp_workshop.server

You can also use the console command:

mcp-workshop

MCP servers in stdio mode use standard input and output to talk to MCP clients, so the terminal will wait instead of showing an interactive prompt. Press Ctrl+C to stop it.

Option 2: HTTP mode

HTTP mode starts the server at http://127.0.0.1:8000/:

mcp-workshop --transport http

You can also use the shorter version:

mcp-workshop --http

Keep this terminal running while Copilot or MCP Inspector is connected.

Connect the server to GitHub Copilot in VS Code

The included .vscode/mcp.json uses stdio mode, which lets VS Code start the server automatically:

{
  "servers": {
    "mcp-workshop": {
      "type": "stdio",
      "command": "${workspaceFolder}/.venv/bin/python",
      "args": ["-m", "mcp_workshop.server", "--transport", "stdio"]
    }
  }
}

On Windows, use the virtual environment Python executable in .venv\\Scripts\\python.exe instead of .venv/bin/python.

If you prefer HTTP mode, start the server first with mcp-workshop --http, then use this server config instead:

{
  "servers": {
    "mcp-workshop": {
      "type": "http",
      "url": "http://127.0.0.1:8000/"
    }
  }
}

Restart VS Code or reload the window if Copilot does not pick up the server right away.

Run MCP Inspector

MCP Inspector lets you check that the server starts and exposes the expected tools:

npx @modelcontextprotocol/inspector .venv/bin/python -m mcp_workshop.server

For HTTP mode, start the server first:

mcp-workshop --http

Then open Inspector and connect to http://127.0.0.1:8000/ using the streamable HTTP transport. The server allows local browser connections from Inspector, so it should not be blocked by CORS.

In the Inspector, look for the hello_world and add_numbers tools.

If Inspector shows a CORS error, stop any old mcp-workshop --http terminal with Ctrl+C, start it again, and make sure Inspector is connecting to http://127.0.0.1:8000/.

Codespaces

If you are using Codespaces, the devcontainer should install the project dependencies and helper tools automatically. It also forwards port 8000 for HTTP mode. After the Codespace finishes setting up, you can run the server, use MCP Inspector, or connect Copilot using the same commands and MCP config above.

The devcontainer installs the GitHub Copilot CLI from npm, which provides the copilot command. Try:

copilot

If copilot is not available, rebuild the container or run:

bash .devcontainer/setup.sh

If you only need to retry the Copilot CLI install, run:

npm install -g @github/copilot

Prompts to try in Copilot

After connecting the server, open Copilot Chat in VS Code and try prompts like:

Use the workshop MCP server to echo this sentence: Hello from MCP.
Use the hello_world tool.
Use the add_numbers tool to add 7 and 5.
What tools are available from the mcp-workshop server?

Python MCP Workshop