Skip to main content
theAuth tracks the two emerging IETF drafts for agent authorization: draft-goswami-agentic-jwt-00 (agentic JWT claims) and draft-liu-agent-operation-authorization-01 (three-layer user-workload-token binding). Claim names are defined in a single file so future audits are a one-file review.

The claim constants

Every claim name lives in packages/core/src/standards/claims.ts as AGENTIC_JWT_CLAIMS. Each constant has a JSDoc reference to the relevant draft section.

Turning claim emission on

Claim emission is off by default and is configured on the two token issuers that support it, not through a single global switch. TheAuthConfig declares a top-level emitAgenticJwtClaims field, but createTheAuth() does not read it, so setting it there has no effect. MCP access tokens. Set emitAgenticJwtClaims: true on the mcp config and supply getAgenticContext, which returns the values to embed. Fields you leave out are omitted rather than invented.
JWT sessions. Set emitAgenticJwtClaims: true in the JWT session config, and pass an agenticContext ({ agentId?, agentType?, trustTier? }) on the SessionUser you create the session for. See JWT sessions. With emission on and a context supplied, a token looks like this (abridged):

What is populated today

Only three claims are ever written, and only from the context you provide: theAuth does not derive these values for you. The trust module is not wired into getAgenticContext, and the TrustTier bands used for the claim (unverified, low, standard, elevated, high) are a separate vocabulary from the trust levels returned by theauth.trust (untrusted, limited, standard, trusted, elevated). Map between them in your own getAgenticContext.

Roadmap

Not implemented, listed so you do not look for them:
  • Three-layer binding (wit, operation): needs workload identity support.
  • act and may_act for RFC 8693 delegation chains.
  • Automatic trust_tier from the trust module.

What this does not aim at

  • OpenID for Verifiable Presentations (OID4VP). Out of scope for agent sign-in. Verifiable Credentials theAuth issues are for audit, not for sign-in.
  • SPIFFE URI scheme. Agent identity uses DIDs (did:key, did:web) which fit the hosted path better.
  • Post-quantum signatures (ML-DSA). Tracked, not scheduled.

MCP OAuth 2.1

PKCE, RFC 9728, RFC 8414, RFC 7591 implementation details.

W3C DID identity

Ed25519 DIDs that back agentic JWT claims like agent_id.

Verifiable credentials

W3C VC issuance for audit export and cross-org trust.

Agent identity

Where agent_id and trust_tier claims originate.
Last modified on October 7, 2026