Claude auto-connect plugins: Claude now vouches for your staff

Claude auto-connect just put Claude in the identity business. On November 3, Anthropic opened a beta that lets Claude Team and Enterprise admins connect approved plugins for every employee at once, with Claude itself signing the identity assertion the plugin trusts. No per-person OAuth, no re-authorizing when tokens expire. It also means your identity provider is no longer the only door, and the offboarding checklist most small teams run just got a gap.

What actually happened

The feature builds on the Enterprise-Managed Authorization (EMA) extension to MCP. Until now, auto-connecting a plugin needed an identity provider that could issue the grant, and per Supabase's docs, Okta was the only one that did at launch. The new option lets Claude act as the issuer instead.

When an eligible member opens a plugin, Claude sends a short-lived signed assertion that the person belongs to your Claude organization. The service checks it and hands back an access token scoped to that person's existing permissions. Admins choose which services to enable and which roles get them. The New Stack counts 14 services at launch, including Supabase, Notion, Figma, HubSpot, Linear, Atlassian, Datadog and Dropbox.

The fine print sits in Supabase's setup guide, and it is worth reading twice:

• When Claude is the issuer, it confirms only Claude org membership and email. It does not check a sign-in at your IdP, your MFA, or Supabase's own sign-in rules.
• Disabling someone in your IdP does not cut off access. You have to remove them from Claude and from Supabase. With SCIM sync, Claude removes a member when your directory deletes them, not when it only suspends them.
• Supabase trusts one issuer URL per SSO provider. Switching to Claude's URL breaks any other AI tool that relied on your IdP-issued grants.
• Tokens already issued live until they expire.

Why it matters for your business

This is a real time-saver. A ten-person shop with Claude, HubSpot and Notion stops losing minutes to consent screens. But the security model moved. If your offboarding is "disable them in Google Workspace," a departed employee who stays in your Claude org keeps a working path into your database tooling. Enforce SSO on your Claude org, even though Anthropic doesn't require it. Add "remove from Claude org" as its own line in offboarding. Start with read-only tools, and give database plugins to the roles that need them.

Sources: Supabase Docs, Enterprise-Managed Authorization for MCP, The New Stack, ClaudeDevs on X.