Anthropic adds centralized MCP connector authorization for Claude Enterprise

Anthropic added enterprise-managed authorization for Claude MCP connectors on June 18, giving Claude Enterprise admins a way to provision connectors once through an identity provider rather than asking each user to approve every tool connection separately. The launch starts with Okta and is available in beta for Claude Team and Enterprise customers. 12

The new feature, called Enterprise-managed auth, lets an administrator connect an identity provider to Claude, choose which MCP connectors are available to the organization, and have employees inherit access when they log in. Anthropic says the access model works across Claude chat, Claude Code, and Cowork. 1

Okta is the first supported identity provider. At launch, Anthropic lists Asana, Atlassian, Canva, Figma, Granola, Linear, and Supabase as supported MCP providers, with Slack coming soon. 1 HubSpot, Ramp, and Webflow are named as early Claude customers rolling the model out across teams. 1

The underlying standard is the Enterprise-Managed Authorization extension to the Model Context Protocol. The MCP project says the extension is now stable and is being adopted by Anthropic, Microsoft, Okta, and a growing set of MCP servers. 3 Okta separately says the beta lets joint customers govern Claude access to participating applications through Cross App Access. 4

This is an enterprise adoption update, not a model release. Its importance is operational: connectors are only useful at scale if companies can manage who connects Claude to Jira, Figma, Supabase, Slack, and other work systems without creating a separate approval trail for every employee.

Anthropic's Help Center says admins can scope connector access by group, team, or role, revoke access by deprovisioning a user in the identity provider, and require a connector to use the corporate identity provider so personal accounts stay out of work tools. 2 The same page also notes that connected services and identity providers remain responsible for their own access decisions, permissions, and data reach. 2

For Claude Enterprise buyers, the move reduces one of the messier parts of deploying AI agents: linking the assistant to real company systems while keeping access auditable. For Anthropic, it also pushes MCP from a developer protocol toward an enterprise distribution layer, where identity providers and software vendors decide how quickly Claude becomes useful inside large organizations.