REAL MCP SERVER · MODEL CONTEXT PROTOCOL OVER HTTP

The vehicle data MCP server for AI agents.

Paste one config block and Claude, Cursor or your internal copilot can search the catalog, read specs, check live recalls and find tire fitment — as native tools, grounded in 1,342,720 real records, with the same token as the REST API.

MCP calls share your plan's API quota — the free tier includes 100 calls/mo, no card.

claude_desktop_config.json
# The whole integration:
{
  "mcpServers": {
    "vehdb": {
      "url": "https://mcp.vehdb.com/mcp",
      "headers": { "Authorization": "Bearer TOKEN" }
    }
  }
}
# Your agent now has:
tools → search_vehicles · get_vehicle · recalls_for_vehicle · tire_sizes_for_vehicle · my_lists

What is the VehDB MCP server?

A hosted bridge between AI agents and a real, enriched vehicle database.

The Model Context Protocol (MCP) is an open standard for connecting AI agents to external tools and data. VehDB runs a hosted MCP server at https://mcp.vehdb.com/mcp that exposes our cleaned, cross-referenced vehicle database — 853,870 cars, 233,450 motorcycles, 84,300 boats and 171,100 tire fitments, enriched with EPA fuel economy and live NHTSA recalls — to any MCP-compatible client.

Instead of scraping pages, pasting spec sheets into a prompt, or hand-writing function-calling schemas, you paste one config block. The agent discovers every tool automatically and starts answering questions with structured, current data — the same data, same token and same quota as the REST API and the dashboard workspace.

Five tools. Real answers.

Typed inputs, JSON output — the model calls them directly, no prompt gymnastics.

search_vehicles

Search cars, motorcycles, boats or tire sizes by make, model, year, body, fuel and drivetrain — the agent resolves any vehicle the user mentions.

"Find me 2024 hybrid SUVs from Toyota."
get_vehicle

Fetch a single record by type and UUID — full specifications plus EPA economy in one call.

"Pull the full spec sheet for that Tucson."
recalls_for_vehicle

Live NHTSA recall campaigns and the NCAP crash-test rating for a vehicle.

"Are there any open recalls on a 2022 Bolt?"
tire_sizes_for_vehicle

OEM and alternate tire sizes that fit a given make/model/year, with computed geometry (±3%).

"What tires fit a 2024 Camry?"
my_lists

The user's saved vehicles — the same lists curated in the dashboard, with notes and part numbers.

"Which vehicles in our lists have open quotes?"

Tool reference

Each tool is typed and self-describing — the agent reads the schema, you don't write one.

Tool Inputs Returns
search_vehicles type, make, model, year, body, fuel_type, drive_type, ev_only, page Paginated matching records with core specs
get_vehicle type, uuid Full specifications + EPA economy fields
recalls_for_vehicle uuid Live NHTSA recall campaigns + NCAP rating
tire_sizes_for_vehicle uuid (or make, model, year) OEM + alternate sizes with computed geometry
my_lists — (uses the token owner) The user's saved lists with notes

The same records are available over the REST API — see the full docs.

An agent grounding its answer in vehicle data

One question, three tool calls, an answer backed by today's records.

claude · vehdb mcp
user: "We're quoting a 2024 Camry. What tires fit it and are there open recalls?"

agent → search_vehicles({ "make": "Toyota", "model": "Camry", "year": 2024 })
       ← { "uuid": "a1b2c3d4-…", "trim": "SE", "fuel_type_name": "GASOLINE" }

agent → tire_sizes_for_vehicle({ "uuid": "a1b2c3d4-…" })
       ← { "oem": "235/45R18", "alternates": ["235/40R19", "245/45R18"] }

agent → recalls_for_vehicle({ "uuid": "a1b2c3d4-…" })
       ← { "open_recalls": 0, "ncap_overall": 5 }

agent: "The 2024 Camry SE runs 235/45R18 from the factory (235/40R19 and
245/45R18 are common alternates). No open NHTSA recalls; NCAP overall 5/5."

Connected in three steps

No SDK, no webhook, no infrastructure. A token and a config block.

1 · Create a token

Grab a Bearer token

Sign up free, open Dashboard → API Tokens, create one. It works for both the REST API and MCP — rotate it anytime.

2 · Add the config

Point your client at the endpoint

Same JSON everywhere — Claude Desktop, Claude Code or Cursor. Snippets below.

3 · Ask your agent

Just talk to it

"What tires fit a 2024 Camry?" — the agent discovers the tools and calls them with live data. No prompting required.

Claude Desktop
// claude_desktop_config.json
{ "mcpServers": { "vehdb": {
  "type": "http",
  "url": "https://mcp.vehdb.com/mcp",
  "headers": { "Authorization":
    "Bearer TOKEN" } } } }
Claude Code
claude mcp add vehdb \
  --transport http \
  https://mcp.vehdb.com/mcp \
  --header "Authorization: \
    Bearer TOKEN"
Cursor
// .cursor/mcp.json
{ "mcpServers": { "vehdb": {
  "url": "https://mcp.vehdb.com/mcp",
  "headers": { "Authorization":
    "Bearer TOKEN" } } } }

Why MCP beats scraping or hand-rolled function calling

The tools, schemas and auth are hosted and discovered — you maintain none of it.

Live, not embedded

Recalls change, model years land, fitments update. A tool call returns today's record — an embedding returns whenever you last indexed.

Structured, not hallucinated

Typed parameters and JSON responses. "Will 245/40R17 fit?" gets answered with computed geometry and live recall records — not a plausible guess.

Zero glue code

No scraper to babysit, no schemas to write, no vector DB to refresh. One hosted endpoint, one token, metered on the same quota as your API.

Approach Setup Freshness Maintenance
MCP server VehDB One config block Live, every call None — hosted & discovered
Hand-rolled function calling Write schema + wiring per tool, per client Live, if you build it You own schemas & updates
Pasted docs / scraping / RAG Build a scraper or index Stale since last index Re-scrape & re-embed forever

Works with your agent stack

Any MCP-compatible client over HTTP — Bearer tokens for you, OAuth2 for AI platforms.

Claude Desktop
Claude Code
Cursor
ChatGPT connector (OAuth2)
Custom MCP agents
Authenticated by default

OAuth2 with dynamic client registration

The MCP server authenticates with the same Sanctum Bearer token as the REST API, metered by the same plan quota. Building a connector for ChatGPT or another AI platform? It also supports OAuth2 via Laravel Passport with dynamic client registration, so providers can onboard your users without manual key exchange — same tools, same quota.

Auth details in the docs

Vehicle Data MCP FAQ

Common questions about the automotive MCP server.

An MCP (Model Context Protocol) server for vehicle data is a hosted endpoint that exposes a vehicle database to AI agents as native tools. Instead of pasting specs into a prompt or hand-writing function-calling glue, an agent like Claude connects to VehDB's MCP server at https://mcp.vehdb.com/mcp and can search vehicles, get specifications, look up live recalls, and find tire fitment directly — with typed inputs and structured JSON output.
Add VehDB to your MCP client config with the HTTP transport, the URL https://mcp.vehdb.com/mcp, and an Authorization Bearer header with your VehDB API token. The agent discovers the tools automatically. Per-client snippets for Claude Desktop, Claude Code and Cursor are above.
Yes. It authenticates with the same Sanctum Bearer token as the REST API and is metered by the same plan quota — the free tier includes 100 calls per month, the Developer plan 100,000. For AI platforms it also supports OAuth2 via Laravel Passport with dynamic client registration.
With MCP the tool definitions, schemas and auth are hosted and discovered automatically — you paste one config block and the agent sees every tool. Hand-rolled function calling means you maintain JSON schemas, wire each tool to the API, refresh them when fields change, and repeat it per client. MCP makes the vehicle database a native capability of the agent with zero glue code.
When a user asks "what tires fit a 2024 Camry and are there any open recalls?", the agent calls search_vehicles to resolve the car, then tire_sizes_for_vehicle and recalls_for_vehicle. It answers from today's live record — computed tire geometry and live NHTSA recall campaigns — instead of guessing from training data.

See the API docs, the REST API, or the vehicle data for business guide.

Give your agent 1,342,720 vehicle records.

Create a free token, paste the MCP config, ask a question. MCP is included on every plan — starting with 100 calls/mo for free.

Operated by Savas OÜ (Tallinn, Estonia — EU) · GDPR-compliant · Billing by Stripe · Terms · Privacy