I made my product database queryable by AI agents - a free API + MCP endpoint

AI assistants are quietly becoming the discovery layer. People don't Google "best power station for a camper van" as much anymore - they ask ChatGPT or Claude. The problem: for a niche like portable power stations, the model usually answers from stale, half-remembered specs and just makes numbers up. I maintain a structured database of 102 portable power stations (specs, $/Wh, cycle life, a transparent score). So I did the obvious thing: I made it queryable by machines and AI agents - free, no auth. A plain JSON API (no key, no signup) # Filter/sort products curl "https://sunsee.cc/api/products.json?max_price=500&sort=value&limit=5" # Get one product's score breakdown curl "https://sunsee.cc/api/score.json?slug=ecoflow-delta-2" # Size a system from a list of appliances curl "https://sunsee.cc/api/calculate.json?appliances=[{"name":"Fridge","watts":60,"hours":24}]®ion=us-south" Every response is clean JSON with consistent fields (battery_capacity_wh , battery_type , dollar_per_wh , sunsee_score , …). No HTML scraping, no inconsistent retailer specs. A discovery endpoint for AI agents The part I actually care about: a single discovery endpoint that describes the available tools, so an agent can find and call them without me hardcoding anything on its side. curl "https://sunsee.cc/api/mcp.json" It returns the tool list (search products, size a system, get a score) with their parameters - the same idea as an MCP tool manifest. Point an agent at it and it can answer "what's the best value station under $500?" or "how big a battery do I need for a fridge + CPAP for 3 days?" with real data. llms.txt for citation guidance I also added a llms.txt (like robots.txt, but for language models) telling models when and how to cite the data: https://sunsee.cc/llms.txt It maps common questions to the right endpoint, and asks for attribution. No idea yet how much models will respect it - but the standard is young and it costs nothing to publish. Why bother? Honestly: if an AI is going to be the thing recommending products, I'd rather it pull transparent, spec-based data (with a public scoring formula, no pay-to-play) than hallucinate. And if it cites the source, that's discovery. Disclosure: the underlying site has affiliate links - that's how I hope to fund it - but the API, the dataset, the calculator, and the scoring formula are all free and open. Try it / break it - API playground starts at https://sunsee.cc/api/products.json - Open dataset (CC-BY-4.0): https://github.com/gkin9/portable-power-stations-dataset - The human version (calculator): https://sunsee.cc/calculator If you build agent/LLM tooling, I'd love to know whether the discovery endpoint is shaped usefully - and what fields you'd want added. Happy to iterate. Top comments (0)