Building AI Agents with MCP
A hands-on multi-module course that builds a personal research assistant from a hand-rolled agent loop up through a deployed, secured production Model Context Protocol server.
This is an O'Reilly Live Training course on the Model Context Protocol, the open standard for connecting AI agents to tools, data and external systems, now under the Linux Foundation's Agentic AI Foundation. Across seven modules it builds one running project, a personal research assistant, adding one layer at a time.
The arc starts from a hand-rolled agent loop with tools as plain Python functions, moves the same tools behind an MCP server reachable from Claude Code and Claude Desktop, shows how the Claude Agent SDK acts as an MCP host, and covers building an MCP server with the mcp-builder skill. Later modules cover a production shape with remote HTTP transport, an authentication seam, intent-grouped tools, hooks and evals, deploying a remote server alongside its first MCP App, and a lab on tool-poisoning attacks and defenses plus multi-server composition.
Every script carries inline dependency metadata for the uv package manager, so exercises run without separate environment setup.