Open Brain
VercelEngineering PostOfficial Source

Vercel Connect now supports TanStack AI

TanStack AI agents can reach OAuth-protected MCP servers through Vercel Connect, with per-user token refresh and consent handled before model execution.

Vercel · Sep 24, 2026
Open Source Open MarkdownOpen JSON
Source Summary

The **@vercel/connect/tanstack-ai** entry point adds **connectMCPTransport**, which attaches Vercel Connect authentication to a TanStack MCP transport and refreshes the token before each request.

Practical Implication

Catch **getConsentChallenge** around createMCPClient and redirect to the supplied consent URL. Handling this before the model runs prevents authorization failures from becoming tool-error text in the conversation.

Agent-Ready Context
The **@vercel/connect/tanstack-ai** entry point adds **connectMCPTransport**, which attaches Vercel Connect authentication to a TanStack MCP transport and refreshes the token before each request.

Catch **getConsentChallenge** around createMCPClient and redirect to the supplied consent URL. Handling this before the model runs prevents authorization failures from becoming tool-error text in the conversation.

The material demonstrates an HTTP Linear MCP client with per-user identity, but does not describe support for other transport types, deployment environments, or failure recovery beyond consent.
Connected Context · Feed7 Judgment

This narrows remote MCP integration from generic connectivity to an explicit per-user authorization boundary: refresh credentials on every request and resolve consent before model execution. That prevents an authentication state from leaking into the agent’s conversational tool-error path. It is a concrete TanStack transport pattern, not yet evidence of portability across transports, deployments, or recovery scenarios.

Expanding Managed Agents in Gemini API: background tasks, remote MCP and moreGemini Managed Agents establishes remote MCP and credential rotation at the managed-runtime level; Vercel Connect supplies a client-side TanStack pattern for per-request refresh plus interactive user consent.ACP: The Universal Remote Control for AI Agents — Alex Hancock, BlockACP carries permission requests between agents and clients, while this integration handles consent before creating the MCP client; together they locate authorization at different boundaries rather than treating it as ordinary tool output.Chat SDK now supports reactions and ephemeral messages on TeamsThe Teams adapter similarly makes private permission flows and custom authentication part of the application boundary, reinforcing that authorization behavior remains integration-specific despite shared SDK abstractions.
Context Map
toolscoding#mcp#agent-sdks#tool-use
Uncertainty
The material demonstrates an HTTP Linear MCP client with per-user identity, but does not describe support for other transport types, deployment environments, or failure recovery beyond consent.