Descope shipped Agentic Identity Hub 3.0 on October 15, extending its identity platform from customer-facing agents into a harder problem: the AI agents and MCP servers employees spin up inside a company without anyone tracking them. The core addition is an Agent Onboarding Portal that turns agent deployment into a request-and-review process instead of a free-for-all. An employee specifies the agent's purpose and the systems it needs to touch, down to individual data sources and MCP servers, and nothing gets provisioned until a reviewer signs off. The problem Descope is targeting is specific: agents that show up with no record of who requested them or who owns them, authenticated with static client secrets that leak into repositories and never get rotated, and often built by someone else entirely — Claude, Cursor, and third-party MCP servers that a security team can't instrument because it doesn't control the code. Version 3.0 addresses the authentication piece by letting an agent sign a short-lived assertion with a private key that never leaves the client, replacing the long-lived API key as the default credential. The release also adds native integration with MCP gateways, including the Agentic AI Foundation's open-source agentgateway project, so a single identity and policy layer can sit in front of every agent and MCP server an organization uses — including outside clients like ChatGPT that the organization didn't build and can't modify. Internal agents get something close to single sign-on: authentication happens once through whatever identity provider the company already runs, and access to each downstream MCP server follows from that without a separate login or consent screen each time, while policy is still evaluated on every request. What's notable here isn't the cryptography — short-lived signed assertions replacing static secrets is a familiar pattern by now. It's the organizational admission embedded in the product: companies have let employees connect arbitrary AI agents to internal systems for long enough that the first fix isn't a better token format, it's a request form. An inventory problem, solved with an approval queue, before anyone gets to the harder question of what those agents should be allowed to do once they're on the books.