Solutions  ·  2026-11-21

Okta for AI Agents extends MCP governance to downstream and third-party hops

SolutionsMedium impactGlobal
On Nov 19, 2026 Okta for AI Agents extended Model Context Protocol (MCP) governance in two ways: internal MCP servers can be registered as Okta OAuth clients so every downstream hop to an enterprise system uses short-lived, task-scoped tokens tied to the requesting user, and Cross App Access (XAA) now extends centralized Okta policy to external XAA-supported MCP servers running their own authorization servers.
It closes the two big production blockers for agent rollouts — overprivileged static credentials behind internal MCP servers and ungoverned shadow AI via third-party MCPs — by carrying user identity and least-privilege policy across every hop, not just the first one. This is a meaningful reinforcement of identity-as-the-control-point for agentic security from a major identity vendor.
Okta customers running or piloting AI agents in production should evaluate the token-per-hop and XAA external-MCP controls now; regulated orgs needing per-user attribution across multistep agent workflows benefit most.
Okta - How to Secure Every Hop in an AI Agent's MCP Workflow
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →