Skip to content

Documentation

oauth.work is one identity layer for an app, its enterprise customers, and the AI agents acting on their behalf. Every tenant is a full OpenID Connect provider on its own host, with its own signing key, and every privileged action lands on an audit trail you can query or stream.

These guides are ordered the way most integrations go: get a token working, then add the parts your customers ask for.

  • Quickstart — from nothing to a working authorization-code flow, with a real tenant and a real token.
  • Authentication — the three ways a caller proves who it is: API keys on the management API, access tokens on the protocol surface, and session cookies on the hosted screens.
  • Errors and rate limits — the error contract, the per-endpoint limits, and the headers that tell you where you are against them.
  • API reference — every endpoint, generated from the same OpenAPI document the platform serves at /openapi.json.
  • MCP authorization — scoped, consented OAuth 2.1 access tokens for MCP servers and the agents that call them, instead of a shared API key.
  • A2A delegation — on-behalf-of tokens between agents, narrowed at every hop.
  • Agent identity — registering agents, per-agent tool scopes, secret rotation, and DID-bound agent credentials.
  • DPoP — sender-constrained tokens, so a stolen token is useless off the client that earned it.
  • OIDC federation — broker a customer’s login to their own identity provider.
  • SAML 2.0 — enterprise single sign-on, including IdP-initiated.
  • SCIM 2.0 — directory sync, so joiners and leavers arrive without a support ticket.
  • Admin Portal — a signed link that lets your customer’s IT admin configure their own SSO, directory sync, and domains.

Need something that isn’t here? Get in touch.