TanStack agents can request MCP consent before calling the model

Vercel Connect now supplies authentication for TanStack AI clients connecting to OAuth-protected MCP servers. Its transport wrapper obtains a fresh token before each request. If the user has not granted access, client creation raises a consent challenge that the application can turn into a redirect before the model runs, instead of exposing the failure only as a tool error.

Vercel · Sep 24, 2026

Why it matters

Handling authorization at the route boundary gives users a clear consent flow when an agent needs another service.

More context

Try the documented TanStack integration with a configured Connect provider and authenticated application user. Catch getConsentChallenge before streaming the model response; the example uses a per-user Linear connection.

Keep in mind

This removes application-managed provider tokens, not the need for correct user identity and permissions. Service access and costs still depend on your setup; the integration was not tested by 1min.

Original read