Car buying MCP
Vehicle facts crossed with a measurement of what AI assistants actually tell car buyers — including where they disagree. Five read-only tools for your own LLM.
Add it to your client
claude mcp add --transport http killer-cars https://killer-tools.vercel.app/api/mcpOr point any MCP client at https://killer-tools.vercel.app/api/mcp over Streamable HTTP. No key, no account, nothing to install.
The tools
find_vehiclesSearch 400 vehicles by class, price, seats, towing, fuel. Rank by rating, price, towing, horsepower or 0-60.compare_vehiclesFull side-by-side facts for specific vehicles.ai_visibilityHow strongly each AI assistant recommends each brand, scored 0-100. Never averaged — you see the split.explain_pickThe actual AI answers behind those scores, verbatim, with the web sources each one cited.list_car_segmentsWhich segments have been measured. Call this first.
Built-in workflows
The server ships four MCP prompts, so your assistant chains the tools in the right order instead of guessing — and knows when to ask you a narrowing question rather than pick for you.
help_me_choose"I have three kids and tow a boat." Finds your buyer profile, then what the assistants recommend for it.what_matters_most"Fuel economy is everything." Resolves your priority onto a measured topic and answers from both datasets.compare_shortlist"Telluride vs Palisade." Resolves the names, compares the facts, adds what AI says about each brand.how_does_ai_see_brand"What does AI say about Kia?" The visibility view — where a brand is named, and where it never comes up.
What the data actually covers
Vehicles: 400 models, 162 classes, 50 makes. Ratings, price bands, horsepower, towing, 0-60, seating, cargo and EPA figures. The source is Car and Driver, and only Car and Driver — this is a single source, not a consensus of reviewers, and the tools say so in every response.
AI visibility: two segments, two assistants. Midsize luxury SUVs, and three-row SUVs under $60K. Each is 60 grounded prompts per assistant across eight topics and seven buyer personas, scored with Claude and GPT, and every answer is kept verbatim with its citations and attributed to the model that gave it.
The scores are never averaged, because the assistants disagree — and the disagreement is the point. In three-row SUVs, Mazda scores 22 with Claude and 61 with GPT. An average would report ~41 and describe a consensus that does not exist. Every brand carries its per-assistant split and the gap between them.
Ask about any other segment and you get not_measured — which means nobody measured it, not that the brands score badly there. Those are completely different answers and the tools keep them apart on purpose.
Where Car and Driver does not state a figure it comes back empty rather than zero. Towing is unstated for 225 of the 400 vehicles. Anything dropped from a filter or a ranking for a missing figure gets named in the response instead of quietly vanishing.
Try asking
- “Which three-row SUV under $60k is best for towing?”
- “What does AI tell people about Kia vs Toyota for family SUVs, and why?”
- “Compare the Telluride and the Palisade on cargo and fuel economy.”
Part of the Human-OS platform