This is the start of a journey, moving from classic software engineering into AI engineering. It begins with the MCP courses from Anthropic and Hugging Face.
I closed the last video with the theory fresh in my head, and zero code of my own written. I could explain what a Tool was, what a Resource was, what a Prompt was. I had no idea if I could actually build an agentic system from scratch.
There is only one way to find out.
So I decided to build something small but complete: a pipeline that listens to GitHub releases, asks Claude to write professional release notes, generates a PDF and delivers everything to a Discord channel. Nothing fancy, just enough to touch every piece of the protocol: server, client, tools, resources, prompts.
I also decided to document every part as if I was teaching it to someone else. Teaching forces me to understand things properly, and I genuinely enjoy doing it. It feels like a challenge in itself. If I cannot explain why tool_use_id matters, that is a sign I have not really understood why it matters.
The result is a README with 20 parts, written as a step by step tutorial that anyone can follow from zero to a working pipeline.
The solution in one paragraph
A GitHub release gets published. A webhook catches it, Claude reads the raw changelog, writes proper release notes, builds a PDF with ReportLab and sends it straight to Discord. No human touches the changelog after it leaves GitHub.
What you will find in the repo
The README is the actual tutorial. 20 parts, each one building on the last: setting up an MCP server, adding tools, testing them with the official Inspector, adding resources and prompts, then building the client side and wiring Claude into a tool use loop. Later parts cover the FastAPI webhook, exposing it with Cloudflare Tunnel, connecting a real GitHub webhook, and finally plugging in the official GitHub MCP server so Claude can read commit history too.
If you are starting with MCP, this is meant to be a warm up project before the more advanced courses.
The nrf52 backstory
This is not my first time building something end to end. Before this project, I spent about two and a half years building a BLE tracking system for restaurant staff, using nrf52 chips. Firmware, hardware, the reporting layer on top, all of it. Different stack, different problem, but the same instinct: build the whole thing, do not stop at the part you already know well.
That project never had a public face. Most of my client work has been under NDA, so there is no portfolio to point to. This release notifier is partly an attempt to fix that. It is also a hint at where I want to go next: combining that hardware background with agentic AI workflows, outside the restaurant world this time. If that overlap sounds interesting to you, I would enjoy hearing about it.
Check it out
If any of this is useful to you, the repo is here: š mcp-release-notifier. The README walks through the whole build, part by part, with screenshots and code at every step.
If you build something from it, break something on purpose, or just have questions, I would genuinely like to hear about it. Comments here, or reach me on LinkedIn or by email at hugoferro(at)gmail.com.
A star on the repo helps more than you would think.
This is post one. More will follow as the journey from software engineer to AI engineer keeps going.
Top comments (0)