Case study · 2026

CarVector.io

Vehicle intelligence, API-delivered. A developer-facing API for vehicle specs, federal recall data, and OBD-II DTC reference — bundled with an open-source Model Context Protocol client and shipped end-to-end inside Geneva Lakes Media.

0+
Vehicles in the catalog (1925–2029)
0+
Federal recall campaigns indexed
0
MCP tools shipped (search, get, recalls, DTC)
0
Distribution channels: site, REST, MCP, npm, registry, GitHub

What we built

Eight disciplines, one shipped product

CarVector is what happens when a brand team, a product team, an API team, and an open-source team are the same team. Every layer below shipped in one pass — no handoffs, no waiting.

Brand identity

"Vehicle intelligence, API-delivered." Dark, developer-facing system: vector-arrow mark, Sora/DM Sans pairing, Void-navy palette with a Vector-cyan accent, JetBrains Mono for code. The brand reads as a tool, not a marketing site.

Site & docs

SvelteKit site spanning home, pricing, data-quality, changelog, and a blog. Full documentation set: quickstart, REST reference, MCP server reference, authentication, errors & rate limits, plans. Built so a developer can land on the homepage and ship in under five minutes.

REST API

Endpoints for vehicle search, single-vehicle details, federal recall campaigns, and OBD-II DTC reference. Bearer / X-API-Key auth, standard X-RateLimit-* headers, per-plan rolling windows, JSON error envelope. Live at api.carvector.io/v1.

MCP server

First-party Model Context Protocol server at api.carvector.io/v1/mcp. Streamable HTTP transport, stateless. Four tools — search_vehicles → get_vehicle → get_recalls → lookup_dtc — designed as a chained workflow an AI agent can reason about.

Open-source MCP client

carvector-mcp on npm: an MIT-licensed, dependency-light Node client. Installed with npx -y carvector-mcp into Claude Desktop, Cursor, Smithery, or any MCP-capable agent. Auth via the user's own API key — no shared secrets, no proxying through us.

Official MCP registry

Published in registry.modelcontextprotocol.io as io.github.carvectorio/carvector-mcp v1.0.1, status active. One of the few automotive-data MCP servers in the canonical registry — listed under the io.github namespace verified by GitHub ownership.

Dashboard, billing, support

Customer dashboard for key minting, usage trend, and plan management. Stripe-backed billing with the full six-tier ladder (Free → Hobby → Developer → Pro → Business → Enterprise). Email-routed support inbox with per-tier triage.

AI-SEO surface

Built for both human SERPs and AI answer engines: llms.txt manifest at the root, six-node JSON-LD graph (Organization, WebSite, SoftwareApplication, WebAPI, SoftwareSourceCode, FAQPage) with sameAs cross-references to GitHub, npm, and the MCP registry, and an explicit robots.txt allow-list for 13 verified AI crawlers.

Where it ships

One product, six surfaces

A developer API is only useful if developers can find it where they already work. CarVector ships simultaneously on the canonical site, the REST endpoint, the MCP HTTP server, the npm registry, the GitHub source, and the official Model Context Protocol registry.

Honest scope

Not the only one. The one built right.

Vehicle-adjacent MCP servers already existed when we shipped — NHTSA wrappers, VIN decoders, recall-only checkers. CarVector's positioning isn't novelty; it's bundling specs + recalls + DTC reference together, honestly framed, with the kind of distribution discipline (npm + GitHub + official registry, day one) that most v0.1 developer products skip. Catalog images are illustrations, not photos. DTC entries are reference data, not a diagnostic tool. Coverage is stated as a range, not an exact count. What you read on the site is what ships.