Table of Contents

1. Challenge

   Introduction to the Model Context Protocol

   Welcome to this Code Lab on building a Model Context Protocol (MCP) server!

   MCP is a standardized specification that allows AI models and agents to interact with external tools and services in a secure and predictable way.

   Imagine an AI chatbot that needs to get the current weather. Instead of having weather-checking logic built into the AI model itself, it can use MCP to ask an external server:

   "Do you have a tool for weather? (Discovery)

   If the server says yes, the AI can then make a second request:

   "Please execute your weather tool for the location 'San Francisco'" (Execution).

   In this lab, you will build the server side of this interaction. You'll create a server that can advertise its tools and execute them when requested. info > This lab experience was developed by the Pluralsight team using Forge, an internally developed AI tool utilizing Gemini technology. All sections were verified by human experts for accuracy prior to publication. For issue reporting, please contact us.

2. Challenge

   Step 2: Setting Up the Basic Server

   First, you need a basic web server to listen for incoming requests.

   You will use FastAPI, a modern, high-performance Python web framework. Your server will expose a single endpoint, /mcp, which will handle all protocol communications via HTTP POST requests.

3. Challenge

   Step 3: Defining MCP Data Structures

   MCP relies on a clear and consistent message structure. You will use Pydantic, a data validation library, to define Python classes that map directly to the JSON structure of MCP requests and responses.

   This ensures that any data sent to your server is valid, and any data you send back is correctly formatted. This is the foundation of the MCP Message Structures concept in action.

4. Challenge

   Step 4: Implementing Discovery

   The first stage of the MCP lifecycle is Tool Registration and Discovery.

   In this phase, an AI agent ask your server what tools it can use. You'll implement a handler for the discovery verb that responds with a list of all available tools — including their names, descriptions, and required parameters.

   This allows the agent to learn how to use your tools dynamically.

5. Challenge

   Step 5: Implementing Tool Execution

   Once an agent has discovered a tool, it can ask your server to run it. This is the Execution phase of the Request/Response Lifecycle.

   In this step, you will implement the logic to handle the execute verb. This involves mapping the requested tool name to the actual Python function, calling it with the arguments provided by the agent, and returning the result in the response.

6. Challenge

   Step 6: Adding Error Handling

   Finally, a production-ready server needs to be robust. You'll add simple error handling to provide a clear response when an agent requests a tool that doesn't exist.

   This is a key Server-Side Implementation Pattern that ensures predictable behavior and helps developers debug interactions with the AI agent.