Vercel Connect GA replaces stored secrets with short-lived runtime tokens

Vercel Connect reaches general availability

Vercel has moved Connect, its framework for wiring applications and AI agents to external services, into general availability on all plans and inside v0, its AI app builder. According to the Vercel blog, the goal is to eliminate stored provider credentials: applications request short-lived tokens scoped to the task at runtime, and those tokens refresh automatically and expire on their own.

Authentication for a token request relies on infrastructure Vercel already provides. Every deployment carries an OIDC identity, and the Connect SDK uses it to prove who is asking, so no extra secret guards the path that issues credentials. Developers register a connector once from the CLI — for instance vercel connect create slack --name acme-slack — attach it to the projects and environments that need it, and call getToken from @vercel/connect only when code actually needs access.

A connector catalog with per-request scoping

The GA release ships with more than 100 preset connectors for services such as Notion and Workday, managed connectors for Slack, GitHub, Linear, Salesforce, Snowflake and Microsoft, plus generic OAuth, API-key authentication and MCP servers. Vercel says custom services follow the same model, so an internal OAuth service registers the same way Slack does.

Scoping happens per request, and identity does too. A token can act as the application, or as a named user after a one-time consent flow defines what it may do on that user's behalf. How narrow a token can get depends on the provider; Vercel highlights GitHub, where a request can be limited to a single repository with read-only permissions instead of a grant that persists across an entire organization.

Built for AI agents

Agent use cases get specific support. The AI SDK adapter lets MCP clients authenticate with tokens issued at runtime, so a tool call to Linear can carry a read-only token tied to one user, with that identity flowing through to downstream actions. GitHub Tools, an open-source tool layer highlighted in the launch post, plugs into Vercel's eve runtime and maps presets like code-review onto Connect scopes, so a token carries only the permissions its toolset requires. Connect is also supported in eve, the Chat SDK and v0, where generated apps can have connectors provisioned at build time with no secret embedded in the output.

Adapters cover the rest of the stack: @vercel/connect/betterauth, @vercel/connect/authjs, @vercel/connect/ai-sdk, @vercel/connect/mcp, @vercel/connect/eve and @vercel/connect/chat. Vercel notes that for eve agents and Chat SDK apps, the two secrets a Slack integration normally keeps — a bot token and a signing secret — vanish from the environment altogether.

Governance arrives with GA

Three capabilities are new at general availability: fine-grained RBAC over who can create and manage connectors, audit logs covering authorization and connector activity, and observability showing how tokens and triggers are used across projects. Combined with per-environment attachment, including Custom Environments, and revocation via a single command, Vercel argues that answering an auditor's question about who had access becomes a log query rather than a cross-project investigation.

Inbound events without webhook secrets

Triggers close the loop in the other direction. Providers send events to Vercel Connect, which verifies signatures server-side, vouches for each event with an OIDC identity and forwards it to the project — even when Deployment Protection is enabled. The application ends up with no bot token to act with and no signing secret to verify webhooks. Vercel quotes Moonpig Group running an internal legal agent on this pattern in production, handling Slack, Jira and Google Drive without managing tokens, secrets or event subscriptions itself.

Pricing

Pricing is usage-based, metered on token requests and trigger events. The Hobby plan includes 500 token requests and 1,000 triggers per month at no extra cost; Pro is billed at $3 per 1,000 token requests and $0.95 per 1,000 triggers, with custom pricing on Enterprise. Teams that used Connect during the beta keep their existing billing terms until September 25, 2026.

Why it matters

Long-lived credentials remain one of the most common ways access leaks, and placing them in a vault makes them harder to steal without limiting the damage once one gets out. AI agents sharpen the problem: they touch more systems, more often, with more autonomy than traditional scripts. Replacing standing secrets with per-request, identity-bound, self-expiring tokens shrinks both the window of exposure and the blast radius of any single leak, while the new RBAC, audit and observability features turn external access into something a team can inspect and revoke. Connect also signals where agent infrastructure is heading — delegated, scoped credentials as a platform primitive rather than something every team builds and rotates for itself.