Give an AI hands — without letting it burn the house down.
An AI model is a genius locked in a glass box: it can talk, but it can't do. The Model Context Protocol (MCP) is how you give it hands — tools it can use, data it can read, actions it can take. This series builds a real, safe, well-tested MCP server one step at a time, for a fictional online candle shop, Bumblewick & Co., whose over-eager AI assistant keeps trying to help in increasingly flammable ways.
Every part ships runnable Python and ends in something you can test — comedy in the story, rigor in the code. Fake candle shop, real MCP skills.
The episodes
- 🕯️ The Glass Box — what MCP is, and your first read-only tool
- ⚙️ Tools, Done Right — schemas, types and validation as guardrails
- 📚 The Shop's Memory — resources and prompts: reading vs doing
- 🪚 Don't Give the Robot a Chainsaw — safety, scoping, confirmation, limits
- 🌍 The Outside World — secrets, real APIs, timeouts and recoverable errors
- ⚖️ Trust, But Verify — testing your MCP server for real
- 🎬 The Grand Unveiling — transports, packaging and shipping it
Every published part appears with its link in the list below — it updates automatically with each new part.
What to expect
Seven parts covering the full journey: the core idea of tools, resources and prompts; schema design and validation; safety for actions that move real money; talking to the messy outside world; testing; and shipping. Each concept is taught with a worked example, a wrong-way/right-way comparison, and hands-on exercises — and it all builds toward one complete, tested server you can clone and run.