Their API, callable from every AI their customers use.
Mapgrid sells geocoding and routing APIs. Their customers started asking for their data inside Claude and AI IDEs — not another dashboard. Instead of building an MCP server, auth, and hosting from scratch, they published their existing API through CMD+K and listed the connect page.
Fictional geodata API company · developer-first · usage-priced
Developers were copy-pasting Mapgrid responses into AI chats by hand. The team estimated a quarter of eng time to build and operate an MCP server properly — OAuth, token storage, rate limits, logs — for a protocol they weren't experts in.
Import the spec, curate the tools
Their OpenAPI spec produced 27 endpoints; they published the 12 that make sense for AI use and rewrote descriptions so models pick the right tool.
12 of 27 endpoints exposedPer-customer identity, not one shared key
Customers connect with their own Mapgrid account via OAuth. Every AI call is attributed, rate-limited, and revocable per customer.
OAuth 2.1 · per-user tokensOne connect page for every AI surface
A hosted connect page — mapgrid's logo, their domain — that works from Claude and any MCP-speaking client. Docs link it; sales demos start with it.
Usage flows into billing
Tool calls are metered per customer and forwarded to Mapgrid's existing usage-based billing, so AI traffic is just another revenue channel.
meter → your billing account“We didn't want to become MCP infrastructure experts. We wanted our API wherever our users' AI happens to be.”
Illustrative scenario — Mapgrid is a fictional company; numbers show the shape of the outcome, not a measured result.