▲ Claude Academy
Sign in

Vault / course/tracks/track-5-mcp-and-tools.md

updated 2026-06-25

Track 5 — MCP & Tools

The Model Context Protocol (MCP) is the open standard for connecting Claude to your tools, data, and systems through a server it can call. Track 3 had you define tools inline; this track teaches you to package capabilities as reusable MCP servers — tools, resources, and prompts — and then to secure, transport, and scale them. By the end you can build an MCP server and wire it into Claude Code or a multi-agent system.

Audience: developers. · Est. time: ~8–10 hours over a week. · Prereq: track-4-claude-code (and the tool-use loop from track-3-building-with-the-api). · Outcome: you can build and secure MCP servers and connect them to Claude.

Sequence

1. Introduction to MCP · ⏱ ~3–4 h + project

  • Course: introduction-to-mcp — what MCP is and why it exists, the client/server model, the three primitives (tools, resources, prompts), transports (stdio vs. HTTP/SSE), and connecting a server to Claude Code via .mcp.json.
  • Project: p12-build-an-mcp-server — build a small MCP server (Python or TypeScript) exposing one tool, one resource, and one prompt, then connect it to Claude Code and use it.

2. MCP: Advanced Topics · ⏱ ~3–4 h

  • Course: mcp-advanced-topics — production concerns: trust and security (what a server can do, approval/config, untrusted servers), authentication, error handling, tool-design principles (naming, granularity, clear schemas, helpful errors), remote/hosted servers, and composing multiple servers.

3. Put it to work in a multi-agent setting · ⏱ ~2 h + project

  • Project: p15-multi-agent-orchestrator — (shared with Track 6) stand up an orchestrator that coordinates worker agents, with at least one worker backed by your MCP server. (If you do Track 6 first, you'll already have this; otherwise it's your bridge into agents.)

You're ready to move on when…

  • You can explain the client/server model and the three MCP primitives, and give an example of each.
  • You've built and run an MCP server and connected it to Claude Code.
  • You can reason about MCP security: what a server is trusted to do, how config gates it, and the risks of untrusted servers.
  • You can state at least four tool-design principles and apply them to your own tool.
  • You've completed p12 (and started p15).

See also