Enterprise-Managed Authorization
Agent SSO, governed on every call
Your identity provider approves each AI agent’s connection with an ID-JAG. Ferentin applies policy, DLP and audit to every MCP call that follows, for every agent.
The problem
Every agent collects its own grants
Under plain MCP OAuth, each agent asks for its own consent and holds its own refresh token for each server. Four agents and five MCP servers make up to 20 grants that security can’t see in one place and the IdP can’t revoke. Even with SSO for agents, the IdP only approves the connection. The tool calls that follow never go back through it.
How it works
How it works
- 1
Okta approves which agents connect
Add Ferentin from the Okta Integration Network, assign the employees who use it and give each AI agent a managed connection to Ferentin. The agent asks Okta for an ID-JAG addressed to Ferentin. If the connection or the assignment is missing, Okta refuses and the agent gets nothing.
- 2
Ferentin redeems the assertion
An Enterprise ID-JAG agent in Ferentin is bound to one identity provider and accepts its assertions only. Each provider starts at Off, so you choose whether it accepts existing users only or provisions new users just in time. Agents without EMA support, such as Cursor, sign in with standard MCP OAuth through the same IdP.
- 3
Policy decides what agents may do
Accepting an ID-JAG says nothing about what its user may do. An MCP policy decides that, by department, identity provider, client and principal type. Acting Principal Type equals Agent matches an AI agent acting for a person, so agents can reach less than the person does at a browser.
- 4
Okta vouches for the upstream too
Add an MCP server such as Linear with IdP-issued Enterprise Managed Authorization. When an agent calls a tool, Ferentin asks Okta for an ID-JAG for that user and redeems it at the server. The agent never holds the upstream credential, and disabling the user in Okta cuts off both hops.
Works with your stack
Works with your stack
Identity providers that issue ID-JAGs, AI agents that request them and MCP servers that accept them. Agents without EMA support still connect through Ferentin with standard MCP OAuth.
Enterprise-Managed Authorization (EMA) is single sign-on for AI agents. Instead of each user clicking a consent screen for each agent and each MCP server, the enterprise identity provider decides whether an agent may connect to a server for that user. It is an MCP extension built on the Identity Assertion JWT Authorization Grant (ID-JAG). Disabling a user at the IdP cuts off the access it granted.
Go deeper
Learn more
- Agent SSO at the front door, governance on every callHow outbound EMA works, including for agents like Cursor with no EMA support of their own.
- Ferentin in the Okta Integration NetworkOkta sign-in for people and Cross App Access for the agents acting for them.
- Glossary: Enterprise-Managed Authorization (ID-JAG)The grant, the three steps and why it moves consent to the identity provider.
See it on your own stack
Walk through it with our team on your identity provider, your models and your MCP servers. Or start free today.