# Manuals-MCP

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Luis Perez](https://cerebralvalley.ai/u/lperezmo)
- **GitHub:** https://github.com/lperezmo/manuals-mcp
- **Demo video:** https://youtu.be/u0MlTWCd8h8
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4027

Every home runs on manuals nobody can find. The dishwasher filter needs cleaning monthly, the car seat has height and weight limits that change as the baby grows, and the washer has an error code that means something specific. That knowledge is printed in a PDF somewhere, so people guess, pay for repairs they could have done, or miss safety limits.

Manuals turns that pile into one place you can ask. Snap the label on any appliance, baby gear or tool. Muse Spark reads the model number, the app searches the web for the right manual, and Muse reads it into what matters: upkeep tasks with intervals, error codes, replaceable parts, specs and car seat limits, every entry tied to its page number. Paper manuals work too: photograph the pages and Muse transcribes them.

Then just ask in plain words, from the website's chat or from any AI app over MCP (the Muse app, Claude, ChatGPT). "Is the car seat cover machine washable?" comes back with the answer and the page it came from. "I just cleaned the dishwasher filter" logs it and moves the next due date. "What's worth doing this week?" shows the few things that matter, safety first. Send a photo of a part and Muse circles it on your own picture.

Trust is built in. Anything the reader wasn't sure about is flagged, and a second model (Jev) checks numbers against the page. Say a fact is wrong and the assistant rereads the page and fixes it, keeping the old version in history. What you say yourself always wins over what the model read.

It is built for real households: shared homes with invites, kids with growth tracking for car seats, a Map of the house by room, system and season, warranty checks from receipts, and catch-up plans for things that were never maintained.

Stack: plain Vercel functions, Supabase (Postgres with row level security, storage, OAuth server for MCP), a remote MCP server with 15 tools, Muse Spark, Muse Image, SAM 3.1 and Jev by TypeSafe.

## More from Meta Global AI Developer Hackathon

- [Crown & Fluff](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4024)
- [AI Symbiont](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4025)
- [marketwingman](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4026)
- [Gr8 AI Solutions](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4028)
- [Zayas House](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4029)
- [Reef](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4030)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4027. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt