REST Connectors: Turn Any HTTP API Into a Tool
Describe an endpoint, its parameters and its authentication — OAuth included — and the model can call it like any other tool.
Describe an HTTP API — its endpoint, its parameters, its authentication — and the assistant can call it like any other tool.
What you describe
The URL and method, the parameters and what they mean, and how to authenticate: an API key, a bearer token, or a full OAuth flow for services that require one. The description of each parameter is what the model reads to decide how to call it, so it is worth writing them the way you would explain them to a person.
Why bother
Because the useful integration is usually specific to you. A ticketing system at work. A home server. A weather API you prefer. A company's internal reporting endpoint. A hobby project's API. None of those will ever be a built-in feature of an AI app, and none of them need to be.
The same rules as every other network tool
Calls are recorded in the network log. Offline mode disables them and the model is told they are unavailable rather than left to fail. Credentials are stored on the phone, in the app's own storage, and are not synced anywhere — because there is nowhere to sync them to.
Related
For services that already speak MCP, use that instead: the tool list comes from the server rather than from your description. And for the reverse direction — other machines using your phone's model — see the LAN API server.
Looking for something else? Every page on this site.