There are two main flows.
1) Agent verified:
The agent provider, like OpenAI, Anthropic, Cursor, or another trusted platform, attests to the user’s identity. The service verifies it and returns credentials.
No email round trip. No human clicking around.
2) User claimed:
The agent registers first, then the user claims it with an OTP. This works even if the agent platform does not support identity assertions yet.
Useful for MCP servers, custom agents, scripts, or anything built directly on top of LLM APIs.