generated: '2026-08-16'
method: searched
status: published
source: https://vehicles.dev/docs#mcp
sources:
  - https://vehicles.dev/docs#mcp
  - https://registry.npmjs.org/vehicles-dev-mcp
deployment:
  mode: local-stdio
  endpoint: null
  install: npx -y vehicles-dev-mcp
  package: https://www.npmjs.com/package/vehicles-dev-mcp
  auth: api-key
  verified: probed
deployment_note: >-
  Local stdio ONLY. The docs describe an intended remote endpoint at https://mcp.vehicles.dev/mcp for
  the Claude API's remote-MCP feature, and state plainly that "the Render service, DNS and TLS must be
  live before clients can use it". Probed 2026-08-16: mcp.vehicles.dev returns DNS NXDOMAIN and no
  HTTPS listener exists, so there is no reachable remote agent surface today. An agent cannot call
  Vehicles.dev without a human first configuring an MCP client to launch the npm package.
server:
  name: vehicles-dev
  transport: stdio
  package: vehicles-dev-mcp
  package_version: 0.1.0
  package_published: '2026-08-14'
  license: MIT
  runtime: Node.js 20+
  url: null
  planned_remote_url: https://mcp.vehicles.dev/mcp
  planned_remote_status: not-resolving
authentication:
  mechanism: environment variable
  variable: VEHICLES_API_KEY
  header_sent: 'Authorization: Bearer <key>'
  key_prefix: vdev_
  note: >-
    The key stays in the server process environment; the model never sees it and the server redacts it
    out of any text a tool returns. A missing key makes the server exit non-zero at launch rather than
    starting and failing every call with a 401.
configuration:
  - variable: VEHICLES_API_KEY
    required: true
    description: Vehicles.dev product API key sent as a bearer credential on every tool call.
  - variable: VEHICLES_API_BASE_URL
    required: false
    default: https://api.vehicles.dev
    description: Origin the /v1/vehicles/ paths resolve against; for pointing at a staging deployment.
  - variable: VEHICLES_API_TIMEOUT_MS
    required: false
    range: 1-600000
    default: "each tool's own budget: 10, 15 or 20 s"
    description: Replaces every per-tool timeout.
clients_documented:
  - {client: Claude Desktop, config: ~/Library/Application Support/Claude/claude_desktop_config.json}
  - {client: Claude Code, config: 'claude mcp add --env VEHICLES_API_KEY=... --transport stdio vehicles-dev -- npx -y vehicles-dev-mcp'}
  - {client: Cursor, config: .cursor/mcp.json}
  - {client: VS Code 1.99+ Agent mode, config: .vscode/mcp.json}
tool_count: 10
tools_schema_note: >-
  Tool names, backing endpoints and descriptions are taken verbatim from the provider's published tool
  table at https://vehicles.dev/docs#mcp-tools. Full inputSchemas were not introspected because the
  server is stdio-only and was not executed; each tool's real input contract is the linked OpenAPI
  operation's parameters (see mcp/vehicles-dev-api-tool-crosswalk.yml).
tools:
  - name: decode_vin
    description: What car is this VIN? Cheapest identity lookup; start here.
    endpoint: GET /v1/vehicles/vin/{vin}
    read_only: true
    source_operation: openapi/vehicles-dev-api-vehicles-api-openapi.yml#getVehicleVinDecode
  - name: get_specifications
    description: What did the factory build? Raw NHTSA vPIC spec sheet — seats, horsepower, GVWR, plant country.
    endpoint: GET /v1/vehicles/specifications/{vin}
    read_only: true
    source_operation: openapi/vehicles-dev-api-vehicles-api-openapi.yml#getVehicleSpecifications
  - name: get_recalls
    description: Has this vehicle been recalled? Every NHTSA campaign for its year/make/model.
    endpoint: GET /v1/vehicles/recalls/{vin}
    read_only: true
    source_operation: openapi/vehicles-dev-api-vehicles-api-openapi.yml#getVehicleRecalls
  - name: get_vehicle_photos
    description: What does it look like? One hero image plus a link to the source listing.
    endpoint: GET /v1/vehicles/photos/{vin}
    read_only: true
    source_operation: openapi/vehicles-dev-api-vehicles-api-openapi.yml#getVehiclePhotos
  - name: search_listings
    description: What is for sale, and what are they asking? A page of live dealer listings plus the total match count.
    endpoint: GET /v1/vehicles/listings
    read_only: true
    source_operation: openapi/vehicles-dev-api-vehicles-api-openapi.yml#getVehicleListings
  - name: get_listing_history
    description: Has this listing dropped in price, and how long has it sat? Scale plan only — other plans get 403.
    endpoint: GET /v1/vehicles/history/{vin}
    read_only: true
    source_operation: openapi/vehicles-dev-api-vehicles-api-openapi.yml#getVehicleHistory
  - name: get_market_value
    description: What is this car worth today? ML-predicted asking price for a year/make/model.
    endpoint: GET /v1/vehicles/market-value
    read_only: true
    source_operation: openapi/vehicles-dev-api-vehicles-api-openapi.yml#getVehicleMarketValue
  - name: get_depreciation
    description: How fast does this model lose value? Model-level retention curve and decay rate.
    endpoint: GET /v1/vehicles/depreciation
    read_only: true
    source_operation: openapi/vehicles-dev-api-vehicles-api-openapi.yml#getVehicleDepreciation
  - name: get_ownership_costs
    description: What does it cost to run? EPA annual and five-year fuel cost, combined MPG, CO2. Fuel only.
    endpoint: GET /v1/vehicles/ownership-costs
    read_only: true
    source_operation: openapi/vehicles-dev-api-vehicles-api-openapi.yml#getVehicleOwnershipCosts
  - name: get_vehicle_report
    description: >-
      Give me everything on this VIN at once: identity, value and depreciation in one call. The most
      expensive operation on the API.
    endpoint: GET /v1/vehicles/report/{vin}
    read_only: true
    source_operation: openapi/vehicles-dev-api-vehicles-api-openapi.yml#getVehicleReport
error_handling: >-
  A non-2xx becomes an isError tool result carrying the HTTP status, the machine `code`, the
  `request_id` and a recovery hint, so an agent can correct a bad argument and retry unaided.
billing_note: >-
  Every tool call is one billable request at the same per-endpoint price and the same successOnly
  rule — charged on 2xx, free on any failure. An exploring agent spends credits faster than a
  hand-written integration, and usage is visible only in the dashboard.