Secure Vercel access for AI agents

Through Vercel, agents can inspect projects and deployments, read build and runtime logs, trigger deploys, check domains, and work with Toolbar comment threads – every call runs through your policies and is logged for audit.

Server URL: https://mcp.vercel.com

Credential modes

Vercel supports bring your own app only. It does not support dynamic client registration, so your organization registers its own integration in Vercel and supplies the client ID and secret. See Credential modes for how the modes compare.

Before you begin

• A Vercel account that can create integrations for your team in the Integrations Console.
• Administrator access to your Agent Authority workspace, to add the resource.

Setup moves between two consoles. You copy a redirect URI in the Agent Authority console, create an integration in Vercel, then return to the console and paste the credentials. Keep both tabs open as you work.

Setup

1. Copy the redirect URI

In the Agent Authority console, go to Resources → Add Resource and select Vercel. Vercel is bring-your-own-app only, so the credentials form opens as soon as you select it, with no mode to pick.

Copy the Redirect URI shown on the form. It includes your tenant's region, so copy it rather than typing it. It takes the form https://oauth.aisecurity.services.<region>.connect.secureauth.com/auth/callback.

Leave this page open. You come back to it in step 3.

2. Register an integration in Vercel

Vercel has no object called an OAuth app. The OAuth client ID and secret come from an integration, so that is what you create.

In the Vercel dashboard, select your account or team from the team switcher, open Integrations in the sidebar, click Integrations Console, then click Create.

Complete the Create Integration form:

• Set Redirect URL to the Redirect URI from the form in the Agent Authority console, using the Copy button next to it. It must use HTTPS.
• Complete the form's remaining required fields.

Save the integration, then open its settings page and scroll to Credentials to copy the Client ID and Secret. Vercel calls these the integration ID and secret.

3. Finish in the console

Back on the Vercel form in the Agent Authority console, paste the Client ID and Client Secret. Leave Token Endpoint Authentication on Auto-detect. Then click Add.

The gateway sends you to Vercel to sign in, through your SSO if configured, and authorize the integration. Review the requested access and click Allow. You return to the console with your Vercel account connected.

Verify the connection

The gateway syncs the Vercel tools automatically. To check the connection end to end, ask your agent to run a request:

List my Vercel projects

If your projects come back, the connection is working.

How users connect

Access is per user. Each additional user connects their own Vercel account the first time their agent calls a Vercel tool: the gateway returns a sign-in link, the user authorizes once, and the tools work from then on. Go to Connections to manage linked accounts.

Vercel's own team and project-level permissions control what each user can reach. The gateway does not widen access beyond what the authenticated user could already do in Vercel.

Authorization goes through Vercel's global OAuth endpoints (vercel.com and api.vercel.com), which are the same for every Vercel account, with no per-team configuration.

Available tools

Required scopes

The gateway requests these four scopes when a user authorizes:

• openid, email, profile – the standard OpenID Connect identity scopes, which identify the signed-in user.
• offline_access – lets the gateway refresh the user's access token without asking them to sign in again.

Policy examples

Rules are evaluated top to bottom and the first match wins. Deny rules only bite when they sit above whatever would otherwise allow the call – including the seeded Allow all rule every org starts with – and allow rules on their own restrict nothing. New rules are inserted at the top of the list, so create them in reverse order or drag them into place. See Policies.

• Read-only access. Deny deploy_to_vercel, import-claude-design-from-url, get_access_to_vercel_url, change_toolbar_thread_resolve_status, reply_to_toolbar_thread, edit_toolbar_message, and add_toolbar_reaction. import-claude-design-from-url writes a design into your Vercel account, and get_access_to_vercel_url mints a link that bypasses authentication on protected deployments – both are easy to miss when scanning for obvious write verbs.
• Deployment visibility without deploys. Allow list_projects, get_project, list_deployments, get_deployment, get_deployment_build_logs, get_runtime_logs, and get_runtime_errors, then deny * below them so nothing else on Vercel falls through to Allow all.
• Block Toolbar write actions. Deny change_toolbar_thread_resolve_status, reply_to_toolbar_thread, edit_toolbar_message, and add_toolbar_reaction.

Next steps

• Create a policy – start from the read-only pattern in Policy examples above.
• Connections – manage the Vercel accounts your users have linked.