MCP server

CarVector ships a first-party Model Context Protocol server. Point Claude, Cursor, ChatGPT, or any MCP-capable client at it and your agent can query real vehicle specs, recalls, and DTC reference natively — no scraping, no glue code.

Endpoint

Streamable HTTP

https://api.carvector.io/v1/mcp

HTTP transport, authenticated with your CarVector API key as a bearer token. Every tool call counts against your plan's rate limit and shows up in your usage, just like a REST request.

Connect a client

The open-source carvector-mcp package is the simplest path — npx fetches it on demand, no install step. Add it to your MCP client's config:

claude / cursor mcp config

{
  "mcpServers": {
    "carvector": {
      "command": "npx",
      "args": ["-y", "carvector-mcp"],
      "env": { "CARVECTOR_API_KEY": "cv_your_key" }
    }
  }
}

Or, if your client supports remote MCP servers, skip the package and point it straight at the hosted endpoint:

remote (HTTP)

{
  "mcpServers": {
    "carvector": {
      "url": "https://api.carvector.io/v1/mcp",
      "headers": { "Authorization": "Bearer cv_your_key" }
    }
  }
}

Keep your key server-side

Your cv_ key carries your plan's full access. Use a desktop/agent client you trust, and never paste it into a public or shared config.

Tools

search_vehicles Find vehicles by year, make, and/or model.

get_vehicle Full specs for one vehicle by id.

get_recalls Federal recall campaigns for a vehicle.

lookup_dtc Diagnostic trouble code reference.

A typical flow: the agent calls search_vehicles to resolve a vehicle id, then get_vehicle or get_recalls. Tool results are the same JSON the REST endpoints return.

Try it

Ask your agent: "What are the open recalls on a 2019 Toyota Tacoma, and what does code P0420 mean?" — it'll chain search_vehicles → get_recalls → lookup_dtc and answer from real data.

← Errors & rate limits Plans & pricing →