Vehicles.dev MCP

Vehicle data, now agent-native.

Give your AI agent typed access to VIN identity, live listings, market value, recalls, specifications, photos, depreciation, ownership costs, and durable vehicle history reports—without writing the integration yourself.

Tools
15 typed tools
Transport
Local stdio
Billing
Successful calls only
Agent session connected
You

What's a 2022 F-150 with 20k miles worth?

01

Request understood

02

get_market_value

03

Grounded answer

MCP callvehicles-dev
{
  "name": "get_market_value",
  "arguments": {
    "make": "Ford",
    "model": "F-150",
    "year": 2022,
    "miles": 20000
  }
}

The agent chooses the tool. Your key stays in the local MCP process.

01 / Quickstart

Up and running in 3 commands

Install the npm package, expose one scoped Vehicles.dev key, and register the server. No source checkout or manual build required. These commands use bash or zsh on macOS and Linux.

  1. 1

    Install

    npm install -g vehicles-dev-mcp

    The published npm package includes the local MCP server. Node.js 20 or newer.

  2. 2

    Authenticate

    vehicles login

    First install the CLI (npm install -g vehicles-dev-cli), then run vehicles login once. It saves your key to ~/.vehicles/credentials.json, which this server reads automatically — so the config below carries no key. On servers or CI, set VEHICLES_API_KEY instead.

    Create a key in the dashboard →
  3. 3

    Connect MCP

    claude mcp add --transport stdio vehicles-dev -- vehicles-dev-mcp

    That is it for Claude Code. Desktop, Cursor, and VS Code use the same server below.

Already use npx? You can skip the global install. See the versionless setup →

02 / How it works

Ask about the car. Not the endpoint.

Every tool carries a narrow input schema, an explicit data boundary, and enough context for an agent to select it from ordinary language. Responses keep the API’s source, coverage, and model assumptions intact.

Follow a complete tool call →
  1. 01

    Ask naturally

    “What is this VIN, what is it worth, and are there recalls?”

  2. 02

    The agent routes the work

    MCP exposes names, descriptions, and schemas so the model can choose and combine the right lookups.

  3. 03

    Vehicles.dev returns the evidence

    Successful tool output preserves the API JSON. Failures include the API request ID and a concrete recovery hint.

03 / Vehicle data catalog

15 vehicle-data and report workflow tools.

Compose focused identity, safety, valuation, ownership-cost, photo, and listings lookups with account-owned report ordering, recovery, polling, and result retrieval. The billable report tool is idempotent and confirmation-gated; every other tool is read-only.

MCP tools

15 account-scoped operations

04 / Clients

One package. Your agent of choice.

The same local server works anywhere that can launch a stdio MCP process. Pick the configuration shape your client expects.

Terminal

Claude Code

Register the global binary from your terminal, then confirm it with /mcp.

claude mcp add --transport stdio vehicles-dev -- vehicles-dev-mcp
JSON

Claude Desktop + Cursor

Both clients accept the same local-server object and fetch the package with npx.

{
  "mcpServers": {
    "vehicles-dev": {
      "command": "npx",
      "args": ["-y", "vehicles-dev-mcp"]
    }
  }
}
JSON

VS Code

Agent mode uses a servers object and an explicit stdio transport.

{
  "servers": {
    "vehicles-dev": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "vehicles-dev-mcp"]
    }
  }
}

Package: vehicles-dev-mcp · MIT licensed · Open the client setup guide →

05 / FAQ

The short answers.

For schemas, environment variables, pricing, and failure handling, the API reference carries the full contract.

Read MCP documentation →
01What does the MCP server do?

Model Context Protocol is a standard way for an AI client to discover and call tools. The Vehicles.dev server turns each supported vehicle-data endpoint into a typed tool, so the agent can choose the operation and fill its inputs from a natural-language request.

02Which AI clients can use it?

Claude Code, Claude Desktop, Cursor, VS Code Agent mode, and any client that can launch a local stdio MCP server. The package runs on Node.js 20 or newer.

03Do I need an API key?

Yes. Vehicles.dev uses one product-scoped vdev_ API key. Run vehicles login to save it to ~/.vehicles/credentials.json — the MCP server reads it automatically — or set VEHICLES_API_KEY for servers and CI. Either way the key is sent directly to the Vehicles.dev API; it is never added to your prompt or returned in tool output.

04What does MCP usage cost?

There is no separate MCP fee. Each successful tool call uses the same plan access, per-endpoint price, rate limit, and success-only billing rule as the REST API. Failed calls are not charged.

05Can an agent change anything?

Only the vehicle-history ordering tool creates anything. It requires a caller-supplied idempotency UUID and explicit confirmation boolean; confirm_order=false makes no request. The other tools are read-only account-scoped lookups.

Ready when your agent is

Give it vehicle context.

Starter is free, includes 1,000 eligible calls each month, and gives you one key for the API and MCP server.