Zero-touch OAuth kills the MCP consent screen, and the maintainers showed up

The friction this kills: a consent screen per server

The announcement by core maintainer Paul Carleton (188 points in the thread) targets the part of MCP everyone hits in an enterprise: every employee individually authorizes every MCP server, security teams get no central policy or audit trail, and personal and corporate accounts get mixed up at the consent screen.

Enterprise-Managed Authorization (EMA) moves the decision into the organization's identity provider. During single sign-on the client obtains an ID-JAG token from the IdP, exchanges it for an access token at the MCP server's authorization server, and the per-server consent screen disappears. Admins define policy once; users automatically get the servers their group membership allows. Okta is the first supported IdP, via its Cross App Access integration. Early adopters include Claude and VS Code as clients and Asana, Atlassian, Figma, Linear and Supabase as servers.

The token format isn't MCP at all

The most useful clarification in the thread came from someone who works on OAuth for a living:

"This is powered by a new token format called an ID-JAG… and isn't MCP specific at all. ID-JAGs can be used for safe and secure data sharing anywhere where data is shared between applications that use the same SSO provider." — an OAuth engineer at Stytch

That reframes EMA as a general enterprise-SSO primitive that MCP happens to be the first big consumer of. An engineer who has fought enterprise OAuth saw the appeal immediately: instead of every app shipping its own half-baked controls for scopes, token expiry and device binding, you run centralized infra that validates user, device, scope and duration for all of them, and he asked for it outside MCP entirely.

The maintainers walked in

The thread filled with the people who shipped it. The Anthropic MCP lead introduced himself directly:

"I am one of the folks at Anthropic that helped deliver this in partnership with Okta and a handful of MCP partners… EMA is now a stable extension." — dend, MCP Lead Maintainer at Anthropic

An engineer on the implementing side surfaced too:

"seeing my own work in the wild is … wild. I implemented the RAS end of this for Atlassian." — the engineer who built the Atlassian side

The Head of Engineering at C1 reported shipping his company as an EMA identity provider the same day, minting the short-lived scoped tokens to "get out of the business of constant OAuth flows" for Linear and other MCPs. One self-described former skeptic conceded the design looked consulted rather than rushed: "you can tell… Anthropic consulted with experts first… rather than vibe coding the spec in isolation."

The catch: convenience the user never sees

The security upside is real: central revocation, a single audit trail, the auth flow pulled out of the agent's context window. One integrations engineer argued that isolating auth outside the model is MCP's actual advantage over plain CLI tools and skills. But the same zero-touch that removes the consent screen also removes the moment of consent. The sharpest reservation named it:

"what I felt bit uncomfortable with is this idea of access being delegated on my behalf by IDP to client without making me aware about it." — a commenter weighing the delegation model

That is the trade EMA encodes: enterprises get governance and employees get fewer clicks, in exchange for the user no longer being the one who grants each connection. For corporate fleets that's a feature; for anyone who liked seeing what they authorized, it's the asterisk.