> For the complete documentation index, see [llms.txt](https://help.cubelet.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.cubelet.ai/platform-and-integration/reference/api-reference.md).

# API reference

*How to discover cubelet.ai's programmatic surface — and why there isn't a separate REST API to reference here.*

## MCP is the API

cubelet.ai's integration surface is Model Context Protocol (MCP) tools, served over HTTP by a federated fleet of servers. There is no separate REST or OpenAPI surface published today — every capability exposed to developers and AI agents, read or write, is exposed as an MCP tool. If you are integrating against cubelet.ai, you are calling MCP tools, not REST endpoints.

## Discovering tools programmatically

MCP is a standard protocol, so capability discovery is standard too: any MCP client can enumerate the tools available on a given server with the standard `tools/list` method, once connected and authenticated to that server. Connect to one of the fleet endpoints described on the MCP servers page, authenticate per that server's requirements, then call `tools/list` to get back each tool's name, description, and a JSON Schema for its inputs — this is generated directly from the server's tool definitions, not a hand-maintained spec, so it always reflects what that server actually accepts. Some tools also declare a JSON Schema for their outputs, per the MCP standard.

Use the schema returned for a tool as the authoritative contract for how to call it: invoke the tool with arguments matching its declared input schema.

For step-by-step client setup, see the Connecting a client pages (Claude, Cursor, ChatGPT).

## No REST/OpenAPI surface

There is no published REST API or OpenAPI specification for cubelet.ai. Don't assume a REST endpoint exists at a URL derived from a tool's name — it doesn't. If your tooling requires an OpenAPI-style spec rather than raw MCP, you'll need an MCP-to-OpenAPI bridge on your side; cubelet.ai does not publish one itself.

## Which server has which tools

The MCP surface is federated: different servers in the fleet expose different tool sets, scoped to different purposes (general cubelet content and simulation tools versus domain-specific consulting/assessment tools). Tool availability therefore depends on which server you connect to. See the MCP servers page for the full server list and what each one is scoped to serve, then use `tools/list` against the specific server you've connected to for the definitive, current tool set.