Why we built Pet Recall Watcher

Every project in the lab starts the same way: with a small, specific frustration that’s annoying enough to be worth fixing, and concrete enough to actually finish.

Pet Recall Watcher started with this one: by the time most people hear about a pet food or product recall, the affected item is already in the bowl or on the shelf. Recall notices exist — they’re just scattered across public agency feeds, manufacturer pages, and the occasional news story. Nobody has time to monitor all of that.

Why MCP was the right shape

We could have built a website that lists recalls. Plenty of those exist. But the more interesting question was: what if you could just ask?

That’s what made MCP the right tool. Instead of building another page for people to remember to check, we built a server that exposes recall data as tools an AI assistant can call. You ask your assistant whether a product is affected; it checks live data and answers. The capability goes to where people already are, instead of asking them to come to us.

What we learned

Scope beats ambition. The temptation was to build a “pet platform.” The useful thing was answering one question well: is this product affected?

Structure is the actual product. Most of the work wasn’t AI — it was turning messy public notices into clean, queryable data the assistant can trust. The model is only as good as what you feed it.

Honest status earns trust. It’s a prototype, and we label it one. It demonstrates a pattern; it isn’t a safety guarantee. Saying that plainly is more credible than pretending otherwise — and it’s the whole reason we use status badges across this site.

See it for yourself

The prototype is live at recalls.agentprolabs.com, and the full write-up is on the project page.

Want notes like this as we build? Join Lab Notes.