Building AI Agents and MCP Tools

You have learned to work with an agent. Now you build one. And the first thing worth knowing is that the loop at the heart of every agent is about twenty lines of code, which means the loop is not the hard part and never was. The hard part is the tools you hand it: what they are called, what their descriptions say, what they return when things go wrong, and what happens when something hostile comes back through one of them.

Building AI Agents and MCP Tools is an honest, practical course about that. You will learn to recognise the many tasks where an agent is the wrong answer and a plain script is the right one, how to write tool descriptions that are effectively prompts and are read as such, what the Model Context Protocol actually is and what problem it solves, and why every single thing a tool returns must be treated as data rather than instruction.

Taught in plain language with clear diagrams and small examples. The security material is defensive throughout: you will learn to recognise prompt injection through tool results so that the agents you build refuse it. Each module blends a short video overview, illustrated lessons, a knowledge-check quiz, and a hands-on interactive activity, including a simulator where hostile text arrives inside an otherwise ordinary tool result.

What you will be able to do:

• Tell the difference between a script, a workflow and an agent, and choose the cheapest one that works.
• Write the four parts of a tool, and understand why its description is a prompt.
• Return errors as data, so the loop can recover instead of guessing.
• Explain what MCP is, what a server exposes, and when a tool should not exist at all.
• Treat every tool result as untrusted input, and recognise prompt injection when it arrives.
• Design one real tool with a name, a schema, a failure mode and a permission boundary.

Who it is for: Builders who have used an agentic tool and now want to extend one or build their own. Best taken after Mastering Claude Code, Security for AI Builders, and Spec-Driven Development.

Course outline: Orientation · What an Agent Really Is · Tools Are the Product · MCP: One Protocol, Any Agent · Trust, Safety and Failure · Capstone: Ship One Tool.