AI Catchup

Claude Makes Enterprise-Managed Auth for MCP Connectors Generally Available

By 4 min read

Anthropic says enterprise-managed auth for MCP connectors is now generally available. Claude Team and Enterprise admins can centralize authorization through their identity provider, while users receive connector access automatically without individual OAuth consent flows.

Anthropic says enterprise-managed auth for MCP connectors is now generally available. The August 24 update names Claude Team and Enterprise admins: organizations can centralize authorization through their identity provider, while users get tools and data connected automatically without authorizing each connector one by one. (ClaudeDevs on X; Anthropic's announcement)

The update is aimed at a familiar enterprise problem: an MCP connector can be useful, but every individual OAuth grant creates another access decision to govern. Enterprise-managed auth moves that decision to the organization’s identity and access layer, so a user’s existing groups and roles can determine which connectors appear in Claude. (Anthropic's announcement)

One admin decision, automatic user access

An administrator provisions a connector for the organization and controls who can use it through the identity provider. Anthropic says access can be scoped by identity-provider group or role, and a deprovisioned user’s connector access can expire through the same central system rather than lingering as an unmanaged personal grant. (Anthropic's announcement)

For the user, the intended experience is simpler: sign in through the organization’s normal single sign-on, then find the connector available in Claude chat, Claude Code, or Cowork when the administrator has enabled it. The announcement describes this as zero-touch setup and says users do not need to authorize each connector individually. (Anthropic's announcement)

Anthropic’s August 24 update says the generally available connector set includes Datadog, Notion, and Slack, in addition to existing support for Asana, Atlassian, Canva, Figma, Granola, Linear, and Supabase. It lists Exa, Miro, and Zoom as coming soon. (Anthropic's announcement)

How the protocol works for connector builders

Anthropic’s documentation describes a signed identity assertion rather than a browser-based consent step. Claude obtains a JSON Web Token issued by the customer’s identity provider, and the connector’s authorization server validates that assertion before returning a normal access token in a back-channel request. (Enterprise Managed Auth documentation)

The documented flow uses the JWT bearer authorization grant and has no browser redirect or per-connector consent page. The authorization server must validate the assertion’s signature, issuer, audience, expiry, and subject, and it must maintain an explicit allowlist of trusted issuer URLs for each tenant. (Enterprise Managed Auth documentation)

From there, the MCP server receives a bearer token in the same general form it would receive after an interactive OAuth flow. Anthropic says the connection can remain active from the user’s perspective while the identity provider’s refresh token remains valid; Anthropic does not control the access-token or assertion lifetime. (Enterprise Managed Auth documentation)

Requirements before adding the flow

A connector cannot simply accept any well-formed JWT. Anthropic’s implementation guide says the MCP server must already implement MCP authorization and Protected Resource Metadata discovery, and the authorization server must recognize Claude’s registered client before the first assertion arrives. (Enterprise Managed Auth documentation)

The authorization server also needs to advertise the JWT bearer grant in its authorization-server metadata. Claude sends a form-encoded token request containing the grant type, signed assertion, registered client ID, and, when the identity provider supports it, a resource indicator tied to the MCP server. (Enterprise Managed Auth documentation)

If a protected tool call returns a 401 Unauthorized response with the documented WWW-Authenticate header, Claude can use the silent enterprise-managed exchange instead of showing the normal interactive Connect card. Authless MCP servers are outside this flow because they never reach the point where an assertion can be exchanged. (Enterprise Managed Auth documentation)

What enterprise teams should take from the update

This is an access-governance update as much as an authentication update:

  1. Put connector eligibility in identity-provider groups or roles.
  2. Enable only the MCP connectors the organization has approved.
  3. Register and allowlist the right identity-provider issuer for each tenant.
  4. Test the JWT bearer exchange and the 401 lazy-auth path before production use.
  5. Keep the identity provider as the place where access is revoked.

The approved announcement establishes general availability and the zero-touch user experience. Anthropic’s implementation guide establishes the protocol requirements, but it does not state additional plan entitlements beyond the Team and Enterprise context named in the announcement. (ClaudeDevs on X; Enterprise Managed Auth documentation)

Sources

Keep building the workspace playbook

Frequently Asked Questions

What changed for Claude MCP connectors?

Anthropic's August 24 update says enterprise-managed auth for MCP connectors is now generally available. Admins can provision connectors through their identity provider, while users receive access automatically without authorizing each connector individually.

Which Claude plans are named in the announcement?

The approved announcement names Claude Team and Enterprise admins. Anthropic's developer documentation describes the protocol and implementation but does not list additional plan entitlements.

What does a connector need to support enterprise-managed auth?

Anthropic's documentation says the MCP server must implement MCP authorization and Protected Resource Metadata discovery, and the authorization server must support the JWT bearer grant and validate the signed identity assertion.

Get the weekly AI Catchup

Tools, practices, and what matters, in your inbox every week.