Multi-agent Communication Protocol Server

MCP (Model Context Protocol) Python Examples

This repository contains examples demonstrating how to build and interact with MCP servers and clients in Python. It covers basic integration, authentication, and an advanced example of a GenAI-powered agent that uses MCP tools.

Features

• Authenticated FastAPI Server: An example of integrating an MCP server into a FastAPI application with bearer token authentication.
• Standalone Tool Server: A standalone MCP server providing weather tools using the National Weather Service (NWS) API.
• Simple Python Client: A basic client for calling a specific tool on the authenticated server.
• GenAI Agent Client: An advanced client that uses Google's Gemini Pro to understand natural language queries and decide which tool to call on the weather server.

Prerequisites

• Python 3.10+
• pip and virtualenv (or pyenv)
• make (for Linux/macOS) or a standard Windows Command Prompt

Setup

1. Clone the repository:
   git clone https://github.com/AgentWorkshop/mcp.git cd mcp
2. Run the setup script: This will create a Python virtual environment in a venv directory and install all the required dependencies.For Linux/macOS:
   make setup
   For Windows:
   run.bat setup
3. Activate the virtual environment: After the setup is complete, you need to activate the environment in your shell.For Linux/macOS:
   source venv/bin/activate
   For Windows Command Prompt:
   venv\Scripts\activate.bat
4. Set up environment variables: The GenAI agent client requires an API key from Google.A file named .env.example is included in the repository. Copy it to a new file named .env.On Linux/macOS:
   cp .env.example .env
   On Windows Command Prompt:
   copy .env.example .env
   Add your Google AI API key to the .env file:
   # .env GENAI_API_KEY="your_google_ai_api_key"

How to Run the Examples

This project provides convenient scripts to simplify running the examples.

• For Linux and macOS users: A Makefile is included. Use commands like make run-remote-server.
• For Windows users: A run.bat script is provided for the Command Prompt. Use commands like run.bat run-remote-server.

The following sections show commands for both environments. The main difference to be aware of is how environment variables are set:

• Windows (cmd.exe): set MCP_CLIENT_TOKEN=my_value
• Linux/macOS/PowerShell: export MCP_CLIENT_TOKEN=my_value

Example 1: Authenticated FastAPI Server (greet tool)

This example demonstrates an MCP server integrated with FastAPI that requires bearer token authentication.

1. Run the Server

The server will start on http://localhost:8000.

2. Run the Client

In a new terminal, first set the required environment variable for the authentication token. For this example, any string will work.

For Linux/macOS:

For Windows Command Prompt:

Expected Output

(Optional) Testing with curl

You can also test the server endpoints directly with curl.

• Unauthenticated request (will fail):
  curl http://localhost:8000/mcp
  Output: {"detail":"Not authenticated"}
• Authenticated request (will succeed):
  curl -H "Authorization: Bearer my-secret-token" http://localhost:8000/mcp
  Output:
  event: endpoint data: /mcp/messages/?session_id=...

Example 2: GenAI-Powered Weather Agent

This example showcases a more advanced use case: a standalone MCP server providing weather tools and a client that uses a Large Language Model (Gemini) to answer natural language questions by calling those tools.

1. Run the Weather Tool Server

This server provides get_forecast and get_alerts tools. It will run on http://localhost:8123.

2. Run the Agent Client

In a new terminal, run the agent client. Make sure you have set your GENAI_API_KEY in the .env file as described in the Setup section.

The client will start an interactive chat loop. The helper scripts run example/client/client.py.

Usage

Once the client is running, you can ask it questions about the weather.

Example Query:

Expected Output:

The client will show the internal reasoning of the model, including the tool call it decides to make. Note that for "Boston", the client code specifically uses hardcoded coordinates.

You can ask for other locations as well, but the model will need to infer the latitude and longitude.

References

• https://medium.com/@miki_45906/how-to-build-mcp-server-with-authentication-in-python-using-fastapi-8777f1556f75