Token exchange advertised but token endpoint accepts unauthenticated clients
grant_types_supported lists urn:ietf:params:oauth:grant-type:token-exchange but token_endpoint_auth_methods_supported is ["none"] — anyone can mint a token for any end-user identity they assert.
What it means
IDENTITY-002 is the companion to IDENTITY-001. IDENTITY-001 fires when an authorization server does NOT advertise RFC 8693 token exchange at all: the agent-identity propagation story is missing. IDENTITY-002 fires when the server DOES advertise it but the token endpoint accepts unauthenticated clients. The feature is present on paper and broken in practice. patchbay reads the authorization server metadata at /.well-known/oauth-authorization-server and checks two fields: `grant_types_supported` must include `urn:ietf:params:oauth:grant-type:token-exchange`, AND `token_endpoint_auth_methods_supported` must be present and contain only `"none"`. When both hold, any party that can reach the token endpoint can perform a token exchange and assert an arbitrary `sub` in the subject_token. There is no cryptographic binding between the agent performing the exchange and the end-user identity it mints a token for. Consequences: - An attacker who discovers the token endpoint can mint access tokens for any user identity the AS is willing to issue, without ever authenticating as a client. - Downstream MCP servers that enforce authorisation based on the `sub` claim treat forged tokens as legitimate. - Audit trails record the attacker's chosen identity, not the attacker. Severity is high: this is a missing authentication boundary, not a missing feature flag. OBO works only if the AS can tell which agent is asking.
How to fix
Require client authentication on the token endpoint. Set `token_endpoint_auth_methods_supported` to one of `client_secret_basic`, `client_secret_post`, `private_key_jwt`, or `tls_client_auth` — and enforce it. Public MCP clients that cannot keep a shared secret should use `tls_client_auth` with per-agent mTLS certificates, or `private_key_jwt` with a per-agent key. Never use `none` for a grant that mints tokens on behalf of a third-party user identity. If your AS cannot distinguish the token-exchange grant from an authorization-code grant for the purpose of client auth, the simplest fix is to require authenticated clients for all grants and migrate public clients to `private_key_jwt`.
Spec
RFC 8693 §2.1 / RFC 8414 §2 (token_endpoint_auth_methods_supported)Run the check against your own server:
npx -y @patchbay-nebius/agent audit <url-or-directory>No install: scan an MCP server in the browser and get a shareable report.
Disagree with this finding on your server? Open an issue with the --json output. False-positive reports are the most useful feedback.