← all tracks
T4
Agents, tools & MCP
How agents get hands. Build your first MCP server, connect real tools to your editor, and make an agent that completes a task end-to-end instead of just answering.
So far the agent has written code. Now it does things — searches, calls APIs, touches your data — because you gave it tools. This is where “AI feature” becomes “AI agent.”
What you’ll learn
- How agents get tools — the mental model for tool use and why it’s the unlock behind every “connect my data” feature.
- MCP (Model Context Protocol) — the standard way to give any agent a tool. Build a server, test it with MCP Inspector, and wire it into your editor. (build a server, MCP for Beginners)
- Designing a good tool — clear inputs, safe defaults, and errors the agent can actually recover from.
- A task-doing agent — search → summarize → deliver, running the loop on its own without you steering each step.
The builds
Your first MCP server (project 07) and a research agent that finishes a real task (10). Done means the agent calls your tool successfully, and completes the task end-to-end unattended.
Lessons in progress. Newsletter subscribers get them first.
Build these in this track
07 Intermediate
Your first MCP server
done → Your editor’s agent can call your tool.
10 Advanced
A research agent that does a real task
done → It completes the task end-to-end without you steering each step.