The Model Context Protocol (MCP) is the open standard AI assistants use to talk to tools — Claude, and any other client that speaks it. Until now, giving your onvibe app an MCP endpoint meant writing the whole JSON-RPC transport by hand and inventing your own way to authenticate callers. Not anymore.
One helper, a full MCP server
With the mountMcp helper you register your tools and their handlers, and the platform takes care of the entire transport for you — initialize, tools/list, tools/call, notifications, batching, CORS and session headers. You write the part that's actually yours (the tools); we handle the boilerplate that's the same for everyone.
OAuth, handled for you
The bigger deal is authentication. Set auth: "oauth" and onvibe becomes an OAuth authorization server for your app, backed by your managed users. A client such as Claude discovers your endpoint, sends the person through an onvibe sign-in, and receives a token scoped to your app and that specific user.
There are no API keys to copy and paste, tokens refresh with rotation, and inside your handlers the signed-in end-user is available through currentUser() — so every tool call is already tied to the right account.
What it feels like
Add https://your-app.onvibe.run/mcp as a connector in your AI app — Claude, or any MCP-capable assistant — log in once, approve access, and your app's tools are live in the chat. Your to-do app can add tasks; your notes app can search notes; your pantry app can tell you what's expiring — all as first-class tools, with the user's identity carried through automatically.