Overview
Pet Recall Watcher is a working prototype of a Model Context Protocol (MCP) server focused on a small, concrete problem: most people only find out about a pet food or product recall after it matters. Notices are spread across public agency feeds, manufacturer announcements, and the occasional news story — and nobody has time to monitor all of them.
Pet Recall Watcher collects that public recall information and exposes it to AI assistants as structured tools. Instead of hunting through websites, a clinic, retailer, or owner can ask their assistant a plain question — “Has anything in this product line been recalled?” — and get a grounded, current answer.
Who it helps
- Pet owners who want a quick, trustworthy way to check whether something they bought is affected.
- Veterinary clinics fielding worried questions and wanting to give accurate, up-to-date guidance.
- Pet retailers who need to pull affected stock quickly and confidently.
- Animal-care businesses — boarding, grooming, rescue — that buy in volume and carry more exposure.
How it works
- Gather. The server ingests publicly available recall notices on a schedule and normalizes them into a consistent, structured shape.
- Expose as tools. Each capability — search by product, check a brand, list recent recalls — is published as an MCP tool that any compatible AI assistant can call.
- Answer in context. Because the assistant calls a live tool rather than guessing from training data, answers reflect current public information and can cite what they are based on.
Why MCP
MCP is an open standard for connecting AI assistants to external tools and data in a structured, permissioned way. Recall data is a good fit: it changes over time, it needs to be accurate, and it benefits from being queried rather than memorized. Building it as an MCP server means the same capability works across assistants that speak the protocol, instead of being locked to one chat product.
Tech stack
Built in TypeScript on Cloudflare Workers, fed by public recall data, and designed to stay within free-tier infrastructure. The live prototype runs at recalls.agentprolabs.com.
Lessons from the lab
- Scope beats ambition. One clear question — is this product affected? — is more useful than a sprawling “pet platform.”
- Structure is the product. Most of the work is turning messy public notices into clean, queryable data the assistant can trust.
- Honest status matters. This is a prototype, and we label it as one. It is a demonstration of the pattern, not a safety guarantee — always confirm critical decisions against the official source.
Roadmap
- Broaden the public sources the watcher monitors.
- Add an optional alert flow (via n8n) so a saved product can notify you on a new recall.
- Publish a short setup guide so others can run their own instance.