PyData Boston 2025

Target Audience

• Python Developers: Engineers who want to make their applications, data, or APIs accessible to AI agents.
• AI Application Builders: Developers creating agents or AI-powered applications who need to connect to external systems. 🤖

Prerequisites

• An intermediate understanding of Python.
• No prior experience with LLMOps is required.

Tutorial Outline (90 Minutes)

Part 1: Concepts & Setup (15 mins)

• What is the Model Context Protocol? We'll explain the MCP Host, Client, and Server architecture.
• A breakdown of the core concepts: JSON-RPC, the initialize handshake for lifecycle management, and the key server primitives (tools, resources, prompts).
• Environment Setup: A quick tour of the provided GitHub repository, explaining the project structure, dependencies, and a simple MCP client we'll use for testing.

Part 2: Hands-On: Lifecycle and Tool Discovery (30 mins)

• Implementing the initialize Handshake and tools/list.
• We'll define a simple Python function and then build the MCP logic to expose it as a "tool," complete with a name, description, and a JSON Schema for its inputs.

Part 3: Hands-On: Tool Execution and Integration (35 mins)

• Implementing tools/call. We will build the endpoint that receives a tool name and arguments from the AI client, securely executes the corresponding Python function, and returns the result in the MCP-specified format.
• We'll run our completed server and use a provided command-line MCP client to simulate an AI Host. Attendees will see their client discover the custom tool via tools/list and then execute it via tools/call, seeing the live JSON-RPC exchange.

Part 4: Recap & The Road Ahead (10 mins)

• Reviewing the key protocol messages (initialize, tools/list, tools/call) and emphasizing how they create a standard, interoperable bridge for AI.
• A brief discussion of next steps, including other primitives like resources, real-time notifications, different transport layers (e.g., stdio), and a final Q&A.