Give every AI agent its own identity and access
A fireside chat with Jiquan Ngiam (MintMCP) and Aaron Parecki (Okta)
Agent identity is the topic of 2026. We're seeing managed agent frameworks released including Claude Tag, where every agent now has it's own identity when you interact with them in Slack. Instead of acting on a borrowed user tokens, agents need their own scoped, admin-controlled account in each tool it touches.
Right now, most AI agents get into your enterprise apps the sketchy way: a user pastes in a long-lived token, clicks through OAuth prompts, and from then on IT has no idea what the agent can touch. Fine for one person and one tool. It breaks the moment a whole company points its agents at Salesforce, Confluence, GitHub, and a dozen other systems.
There's now a standards-based fix, and this conversation digs into it.
What we'll cover
✅ Enterprise-Managed Authorization: granting, scoping, revoking agent access
✅ Why connected MCP servers work with zero manual setup and no consent screens
✅ How XAA kills messy per-app auth flows and improves security and observability
✅ Making it all work across Claude, Cursor, ChatGPT, and internal agents through your IdP
You rarely get to tighten security and reduce friction in the same move. This is one of those times.
Your speakers
Aaron Parecki, Director of Identity Standards at Okta. Maintains oauth.net, authored "OAuth 2.0 Simplified," and edits the IETF OAuth Working Group defining XAA.
Jiquan Ngiam, CEO & Co-Founder of MintMCP. Co-author of "Securing the Model Context Protocol" and runs 20+ agents in production on MintMCP.
MintMCP is the governance layer for MCP infrastructure: which servers run, who can access what, full audit visibility.
Okta is the identity layer for the enterprise, now extending those controls to AI through Cross App Access.
By submitting your information, you agree to share your data with and be contacted by MintMCP about their services. Your information will be processed in accordance with our privacy policy.