Anthropic just solved one of enterprise AI adoption's biggest bottlenecks: the fragmentation of OAuth authorizations needed to connect Claude to workplace tools. Today, the company launched Enterprise-Managed Authorization (EMA), an open extension to the Model Context Protocol (MCP) that delegates access management to the corporate identity provider. In practice, an IT admin configures connectors once in Okta, and every employee automatically inherits those permissions on their first Claude login — no forms, no individual OAuth flows, no support tickets.
What Did Anthropic Launch with Enterprise-Managed Authorization?
EMA is a formal extension to the MCP specification, published as a draft in the ext-auth GitHub repository (based on SEP-990) and implementing the Identity Assertion JWT Authorization Grant (ID-JAG) standard. The flow is elegant: when an employee logs in via corporate SSO, their identity provider (currently Okta using its Cross App Access protocol) issues an authorization JWT that MCP servers exchange directly for an access token — no extra screens, no authorization clicks. The seven connectors supported at launch are Asana, Atlassian (Jira and Confluence), Canva, Figma, Granola, Linear, and Supabase, with Slack coming next. On the client side, EMA works in Claude (chat), Claude Code, and Cowork for Team and Enterprise plans, and in VS Code starting with version 1.123 in preview mode.
"Anthropic's enterprise MCP authentication changes the AI adoption model for businesses: instead of every employee configuring their own integrations, IT sets it up once and AI arrives ready to work from day one. That's exactly what an SMB needs to scale."
Davarion Group & LabsReal Impact for SMBs
- 01Frictionless mass deployment: Ramp onboarded 2,000 employees with zero additional steps — the enterprise AI adoption standard shifts from days to minutes.
- 02Automatic and secure deprovisioning: when an employee leaves, losing Okta access automatically cuts their access to all Claude MCP connectors, eliminating the risk of orphaned accounts with access to Asana, Jira, or Supabase.
- 03Open standard: any SaaS tool can implement EMA using the public specification, meaning the ecosystem of secure connectors will grow rapidly beyond the initial 7.
- 04Immediate action if you use Claude Team or Enterprise: activate enterprise authentication in your admin console and connect your Okta tenant — your employees will have Asana, Atlassian, Figma, and Linear available in Claude from their next login.
For small and medium businesses already using Claude in their operations, this update represents a qualitative leap in AI governance. Until today, deploying Claude with external tool connectors required each user to complete individual OAuth flows for each application — generating adoption friction and security risks from long-lived tokens without centralized oversight. With EMA, the IT department maintains full visibility and control from the same panel where it manages all other corporate access. Tom Moor, Head of Engineering at Linear, put it this way: 'Logging in once and automatically having all your MCP connectors set up is pretty magical.' The integration with VS Code 1.123 also positions Claude Code as the enterprise coding agent with the best enterprise integration on the market, just as the battle for AI coding agents intensifies.
At Davarion Group & Labs, we help businesses in Houston, TX and across Latin America implement autonomous AI agents that integrate with the tools they already use: Asana, Atlassian, Slack, databases, and more. With Anthropic's Enterprise-Managed Authorization launch, the path to deploying a configured, secure, day-one-ready Claude agent has shortened dramatically. If your company uses Claude Team or Enterprise and you want to maximize this new capability, the Davarion team can help you design the connector architecture, configure access policies in Okta, and train your team to use Claude with native access to all their work tools. Visit davarion.com to schedule a consultation.
