Module 18
Model Context Protocol (MCP) and Agent SDK
The Model Context Protocol turns the tangle of one-off integrations between AI apps and tools into one standard plug. You learn how hosts, clients and servers talk over JSON-RPC, build and test your own MCP server and connect it to Claude Desktop and Claude Code, then assemble teams of agents with hosted tools, function tools, agents as tools and handoffs. The module ends with a CodeRabbit-style AI reviewer that comments on every pull request.
By the end of this module you can
- explain the N × M integration problem, what MCP standardises, and where the USB-C analogy breaks.
- describe hosts, clients and servers, tools, resources and prompts, how versions and capabilities are negotiated over JSON-RPC, and the stdio and Streamable HTTP transports.
- build an MCP server with FastMCP, test it in MCP Inspector, and connect it to Claude Desktop, Claude Code and your own Python client.
- build a multi-agent system with the OpenAI Agents SDK using hosted tools, function tools, agents as tools, handoffs and MCP servers, and compare it with the Claude Agent SDK.
- ship an AI code reviewer that posts structured, deduplicated inline comments on pull requests within a cost budget.
Lessons
5 in this module- 01What is MCP and why it mattersExplain the N × M integration problem, what MCP servers and clients do, how MCP relates to tool calling, and where the USB-C analogy breaks.
- 02MCP architectureTrace an MCP session message by message, explain hosts, clients, servers and their primitives, and choose between the stdio and Streamable HTTP transports.
- 03Build a tiny MCP serverBuild an MCP server with two tools, a resource and a prompt using FastMCP, test it in MCP Inspector, connect it to Claude Desktop and Claude Code, and call it from Python.
- 04Agent SDKs: hosted tools, function tools, agents as toolsBuild a team of agents with the OpenAI Agents SDK using hosted tools, function tools, agents as tools, handoffs and an MCP server, and compare it with the Claude Agent SDK.
- 05Project: AI code reviewer like CodeRabbitShip a GitHub Action that reviews every pull request with RAG over your style guide, structured findings from Claude, validated and deduplicated inline comments, and a cost budget.