MCP Project Setup

This document outlines the steps to set up the `mcp` project environment.

1. Create Conda Environment

Create a new conda environment named `mcp` with Python 3.12:

conda create -n mcp python=3.12 -y

2. Install uv

Install the `uv` package manager using the following command:

curl -LsSf https://astral.sh/uv/install.sh | sh

3. Install fastmcp

Install the `fastmcp` package using `uv` in the `mcp` environment:

conda run -n mcp uv pip install fastmcp

4. Verify Installation

Confirm the `fastmcp` installation by running the following command:

conda run -n mcp fastmcp version

5. Running the Hello World Server

`mcp_hello.py` is a "hello world" type mcp server. You can run the MCP inspector for it using the following command:

conda run -n mcp fastmcp dev mcp_hello.py:mcp

6. Connecting with Proxy Session Token

Copy the provided session token from CLI, click on the provided link, paste in Configuration -> Proxy Session Token, click connect.

7. Inspect hello_world tool

Click on Tools in the top menu bar. "hello_world" should be listed with a parameter "name". Input your name and click "Run Tool". The tool should succeed and return a greeting.

8. Running Resource Tests

`mcp_resources.py` defines MCP resources. You can run tests for these resources using the `--test` argument:

uv run mcp_resources.py --test

9. Weather Server

`mcp_weather.py` exposes a tool to get current weather data from OpenWeatherMap.

Before running: Ensure you have set your `OPENWEATHER_API_KEY` in the `.env` file:

OPENWEATHER_API_KEY=YOUR_API_KEY_HERE

To run the weather server manually:

conda run -n mcp fastmcp dev mcp_weather.py:mcp

To run tests for the weather tool:

uv run mcp_weather.py --test

10. Integrating with Gemini CLI

To allow the Gemini CLI to automatically start and connect to your `mcp_weather` server, you need to configure its `settings.json` file.

1. Locate `settings.json`:

The `settings.json` file is typically located at:

If the file or directory does not exist, create them.

2. Add `mcpServers` entry:

Add the following entry to the `mcpServers` section in your `settings.json` file. Replace `/mnt/d/Projects/_sandbox/mcp/` with the absolute path to your `mcp` project directory.

{
      "mcpServers": {
        "weather_server": {
          "command": "uv",
          "args": [
            "run",
            "/mnt/d/Projects/_sandbox/mcp/mcp_weather.py"
          ],
          "cwd": "/mnt/d/Projects/_sandbox/mcp",
          "timeout": 10000
        }
      }
    }

Once configured, when you run `gemini`, the CLI will automatically start your `mcp_weather.py` server and make its `get_current_weather` tool available to the Gemini model.

References

• Gemini CLI Configuration: https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/configuration.md - For information on setting up MCP with the Gemini CLI.
• FastMCP: https://github.com/jlowin/fastmcp - The FastMCP library used for building MCP servers.