MCP server · ChatGPT app · REST API

Vehicle data tools for AI agents

Decode VINs, check recalls, pull verified buyer pricing — free, no key. And the thing agents can't get anywhere else: the original factory window sticker, fetched by VIN and parsed to JSON — every option, every price, the real MSRP.

Hosted MCP server

Claude Desktop, Claude Code, Cursor, VS Code, any MCP client

https://www.carwhere.com/api/mcp

Free tools, no key: decode_vin, check_recalls, get_pricing, validate_vin_check_digit — plus paid-report checkout tools your agent can hand the user.

ChatGPT app

ChatGPT (live in the app directory)

chatgpt.com/plugins/plugin_asdk_app_6ac56d4d2bc8819193ee5cd57fd19b8c

decode_vin and check_window_sticker inside ChatGPT — stickers come back as a free PDF link.

Local MCP via npm

Anything that runs a stdio MCP server

npx -y carwhere-vin-mcp

Same free tools, running locally. One line in Claude: claude mcp add carwhere-vin -- npx -y carwhere-vin-mcp

CLI & SDK

Terminals, cron jobs, your own agent’s tools

npm install -g carwhere

carwhere decode <vin> free from the terminal; parse-sticker, specs, recalls, values on your key. Same package imports as an SDK: import { decodeVin, parseSticker } from 'carwhere'.

REST API (production)

Your own agent’s tools, backends, pipelines

x-api-key — $49/mo, 1,000 calls, 7-day trial

Six endpoints incl. the two agents can’t get anywhere else: original window-sticker PDFs and parse-sticker (the Monroney as structured JSON).

Two tool calls, start to finish

A real transcript — the decode is live NHTSA data and the parse is the actual response for this VIN:

You: I'm looking at a used Sierra, VIN 1GTPHGE80TZ138393.
     What did it cost new, and with what options?

tools/call decode_vin        { "vin": "1GTPHGE80TZ138393" }
→ 2026 GMC Sierra Denali · 3.0L I6 Diesel · Fort Wayne, IN · 0 open recalls

tools/call parse-sticker     GET /api/dealer/v1/parse-sticker?vin=…
→ {
    "vehicle": "2026 GMC SIERRA 1500 DENALI",
    "baseMsrp": 65900,
    "destinationCharge": 2595,
    "totalMsrp": 72055,
    "options": [
      { "name": "TECHNOLOGY PACKAGE", "price": 1275 },
      { "name": "POWER SUNROOF", "price": 995 },
      { "name": "ENG: DURAMAX 3.0L TURBO-DIESEL", "price": 795 },
      { "name": "STERLING METALLIC", "price": 495 }
    ],
    "confidence": "high"
  }

Agent: It stickered at $72,055 — a $65,900 base Denali with the
Technology Package, sunroof, and the Duramax diesel. Here's how
the asking price compares to that…

What agents build with it

Pre-purchase check

decode_vin + check_recalls

Agent verifies the listing describes the car the VIN actually is — trim, engine, drivetrain — and flags open safety recalls, including do-not-drive campaigns.

Original build & MSRP

parse-sticker

The question agents get constantly — "what options does this car have, what did it cost new?" — answered from the original factory document, as JSON. No other API does this.

Is this a fair price?

get_pricing

Verified buyer-paid transactions for the year/make/model — what people actually paid, not list prices — so the agent can anchor a negotiation.

Fleet recall sweep

REST /dealer/v1/recalls

One brand-level call returns current campaigns newest-first with severity flags — schedule it and diff for new campaigns across a whole fleet.

Listing writer

decode_vin + parse-sticker

Feed a VIN, get the factory options with original prices — the agent writes a marketplace listing that proves the car’s equipment instead of guessing it.

Sell a report, in-chat

buy_vin_report / buy_used_car_report

When the user needs title/history depth, the MCP server returns a Stripe checkout link your agent relays — the user pays CarWhere directly; no fulfillment on your side.

Going to production?

The free MCP tools are for agents acting for one user at a time. When your agent is a product — fetching stickers, parsing Monroneys, sweeping recalls at volume — the REST key is the lane: six endpoints, $49/month, 1,000 calls, 7-day trial, key emailed at checkout. Working code for every endpoint is in the API guides.

Get an API key MCP setup guide

FAQ

Is there an MCP server for car data?

Yes — CarWhere hosts one at https://www.carwhere.com/api/mcp (streamable HTTP, works with Claude, Cursor, VS Code, and any MCP client), and ships the same free tools as a local stdio server via npm (npx -y carwhere-vin-mcp). Free tools need no API key: VIN decode via NHTSA vPIC, open-recall checks, verified buyer pricing, and offline VIN check-digit validation.

What vehicle data can an AI agent get that it can’t get elsewhere?

Original window stickers. For 12 brands (GM, Ford/Lincoln, Stellantis, Hyundai, Genesis), CarWhere fetches the factory Monroney by VIN and — on the production API — parses it into structured JSON: base MSRP, destination charge, total MSRP, and every factory option with its price. That answers "what options does this car have and what did it cost new" from the primary document; no other provider offers it.

Does the agent need an API key?

Not for the hosted MCP’s free tools or the ChatGPT app. The production REST API (window-sticker fetch, parse-sticker, specs, recall feeds, market values) uses an x-api-key — $49/month for 1,000 calls with a 7-day free trial, key emailed at checkout.

What are the rate limits?

The hosted MCP allows 120 requests/hour per IP for the free tools. The production API meters 1,000 calls/month per key (parse-sticker counts as 5; failed parses and unavailable VINs are never charged).

How should my agent cite the data?

Decode and recall data: "CarWhere via NHTSA vPIC" with the permalink https://www.carwhere.com/vin/{vin}. Sticker data: the original factory window sticker, fetched and parsed by CarWhere. The MCP tools return ready-to-relay links so users can see the full record.