OAuth 2.0 client credentials for machine-to-machine access

What it is

RFC 6749 section 4.4 defines the client credentials grant: the client posts grant_type=client_credentials with its credentials (HTTP Basic or form parameters) to the token endpoint and receives an access token, optionally limited by scope. RFC 8707 adds the resource parameter so the client says which protected resource the token is for, letting the server issue audience-bound tokens.

Why it matters

Agents and background services need to call APIs without a human present. Reusing a user's credentials or long-lived static keys in every request is worse than short-lived tokens bound to a resource and scope.

How to apply

• Register each service or agent as its own client; never share a client secret between deployments.
• Request only the scopes needed and pass resource when the server supports it; verify that the server enforces the audience.
• Cache the token until shortly before expires_in; handle invalid_client and invalid_scope errors without retry loops.
• On the server: store secrets hashed, rate-limit the token endpoint, issue short-lived tokens, log issuance with client ID.
• Serve authorization server metadata (RFC 8414) so clients can discover the token endpoint.

Pitfalls

Sending credentials in the query string. Treating the access token as an identity of a person. Forgetting that "no user" also means no consent screen: the client's permissions are the operator's decision.