> 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/connecting-a-client.md).

# Connecting a client

*How to connect any MCP client to a cubelet.ai server, from picking the right endpoint to authenticating.*

## 1. Pick a server

cubelet.ai's developer surface isn't a single endpoint — it's a federated fleet of MCP servers, each scoped to a different part of the platform (cubelet content and simulations, a specific domain's tools, the Scaffold diagnostic, and so on). Start at the MCP servers page to find the server that has the tools you need.

## 2. Connect over streamable HTTP

Every server in the fleet speaks MCP over streamable HTTP at its own `/mcp` path — for example `https://manufacture.cubelet.ai/mcp`. Point your MCP client at that URL. The client-side setup is the same shape regardless of which server you're connecting to; only the URL and the tools it exposes change.

## 3. Authenticate

Fleet servers require a platform-issued bearer token on every request. See Auth & access for how to obtain one and which header to send.

## 4. Follow your client's quickstart

* **Claude** — Claude Code CLI; note the connector limitation for Claude Desktop and claude.ai.
* **Cursor** — `.cursor/mcp.json`.
* **ChatGPT** — Developer Mode connectors. Read the auth limitation on that page before you start; it currently rules out bearer-token-gated servers.

Each quickstart assumes you already have a credential from Auth & access and know the `/mcp` URL of the server you need from the MCP servers page.