aiengineering.guideaiengineering.guide

Working SWE → AI engineer · Stop 9 of 15 lessons

MCP, in the shape you'll use it

What this lesson covers

MCP — the Model Context Protocol — is the standard that turns “here’s a tool I wrote” into “here’s a server anyone’s agent can talk to”. Whether it’s the right tool for your job is a real decision, not an assumption.

The outline

  1. What MCP is. A JSON-RPC protocol for exposing tools, resources, and prompts to any model client that speaks it.
  2. The three primitives. Tools, resources, prompts — what each is for and why keeping them separate matters.
  3. Client vs. server. Which side you write, when, and why the split matters for security.
  4. Connecting an existing server. Filesystem, GitHub, Slack — the already-written servers you can plug into today.
  5. Writing your own server. The minimum Python to expose a function as an MCP tool other agents can call.
  6. When MCP is overkill. Cases where a plain function inside your agent is the right answer.

Coming soon

In outline.

Outline

Enter to go · Esc to close