Review · November 16, 2026
OAuth Client ID Metadata Document (CIMD)
The Problem
An OAuth client needs a client_id and registered metadata at every authorization server it talks to. A client that talks to servers it has no relationship with can't register manually.
Dynamic Client Registration lets the client register itself, but each server then has registration data to manage and clean up.
CIMD turns it around. The client_id is an HTTPS URL. The client publishes its own metadata as JSON at that URL, and the authorization server fetches it.
How It Fits In
This is an adopted working group draft, not a new submission. It is at -03 and squarely in charter.
It reuses the RFC 7591 client metadata vocabulary and adds one authorization server metadata property, client_id_metadata_document_supported. No new endpoints. No new roles.
The new relationship is that the authorization server makes outbound HTTP requests to a URL the client chose. Most of the security text exists because of that.
Other work already leans on it. The OAuth 2.1 draft (-17) mentions it in its consent guidance, and dozens of other drafts reference it.
So what is left before publication? The -03 revision is a large one. The text grew from about 20 pages to 30, and it contains new normative requirements. It still has an inline editor's note in Section 5 that waits on OAuth 2.1 issue 266. This is not ready for working group last call until the items below are settled.
Summary
- The client picks a Client Identifier URL. It must be
https, with a path other than/, no fragment, no userinfo, and no dot segments (Section 3). - The document at that URL is JSON with a
client_idproperty that exactly matches the URL (Section 4). Comparison is simple string comparison. The server MUST NOT normalize. - The authorization server fetches the document, without following redirects, and treats anything other than a 200 as an error (Section 6).
- Shared secrets are banned (Section 4.1). A client that wants to authenticate publishes public keys via
jwksorjwks_uriand uses a method likeprivate_key_jwt(Section 5).
A server can also pre-register a Client Identifier URL and skip the fetch at request time (Section 9.2). The draft expects this to be common in enterprise deployments.
What changed in -03
- Client authentication moved out of Security Considerations into a normative Section 5.
- Clients SHOULD use the plural
token_endpoint_auth_methods_supportedfrom OpenID Connect RP Metadata Choices. The singulartoken_endpoint_auth_methodis now NOT RECOMMENDED. - The RFC 7591 default of
client_secret_basicno longer applies. A document with neither property meansnone. - Section 6.1 separates transient failures from terminal ones, and defines the errors at the token and PAR endpoints.
- Prefetching and re-serving
logo_uriis now a MUST if the logo is displayed (Section 10.9). - The
client_id_expires_atTBD from -02 is gone.
Three additions are not in the Document History. Section 7 requires the iss authorization response parameter from RFC 9207. Section 10.1 covers a Client Identifier URL that equals an issuer identifier. Section 10.12 requires persisted authorization decisions to be scoped to the client and resource pair. The first and third are new MUSTs on every authorization server.
Interoperability
The core is tight. The URL rules, the exact-match client_id check, and the no-redirect fetch are all MUSTs. Two implementations will agree on what a valid document is.
Significant interoperability concerns
Section 8 contradicts itself. Servers that publish metadata "MUST include" client_id_metadata_document_supported, and the property is then defined as "OPTIONAL". Clients use this flag to avoid sending users to a dead end. Pick one.
The client can't always tell which authentication method was selected. A client can list ["none", "private_key_jwt"]. The server picks one it supports. The client learns the answer from the server's token_endpoint_auth_methods_supported metadata, which is only a SHOULD (Section 5). A client that guesses wrong gets invalid_client.
A normative dependency on an OpenID Foundation spec. Section 5 says servers "MUST recognise" token_endpoint_auth_methods_supported as a client metadata property, defined in RP Metadata Choices. It has the same name as the RFC 8414 server metadata property.
Errors at the authorization endpoint are underspecified. Section 5 says to abort the authorization request "with invalid_client". Section 6.1 says the server MUST NOT redirect to the client. So the error code never reaches the client. The editor's note admits the underlying text doesn't support this yet.
Defaults for grant_types and response_types are not addressed. Section 6 requires the server to intersect these with what it supports. Section 5 overrides the RFC 7591 default for the authentication method. Nothing says whether the RFC 7591 defaults apply when these two are omitted.
Caching has no numbers. Minimum and maximum cache durations are SHOULDs with no values (Section 6.2). A client that rotates keys can't predict when a server will notice.
Mechanisms that could be removed
The CIMD Service material appears twice, in Section 10.13 and Appendix A. The appendix calls itself non-normative and then uses MAY, RECOMMENDED, and MUST. One short non-normative section would do.
Section 10.12 is about resource indicators, not client identifiers. It is a general OAuth requirement that may belong in a different document.
Writing Quality
Clear and conventional IETF style. The -03 additions are less polished than the rest:
- Section 5 uses a lowercase "must" for a required error ("the Authorization Request must return an error of invalid_client").
- Section 4.1 says
jwksandjwks_uriMUST NOT both be present "in the same request or response". That wording is from RFC 7591. Here there is only a document. - Section 4 says the document "MAY also be served with more specific content types". No base content type is stated.
- Section 4.4 protects only
client_urifrom same-origin rules. The other informative URIs are not mentioned. - Section 10.12 says "resource" without citing RFC 8707.
Authors
Aaron Parecki (Okta). 2 published RFCs: RFC 9728 (Protected Resource Metadata) and RFC 10017 (OAuth 2.0 for Browser-Based Applications). 18 active Internet-Drafts, including OAuth 2.1, CIMD and ID-JAG. An established author doing a lot of active writing.
Emelia Smith (independent). No published RFCs per her Datatracker profile. 3 active Internet-Drafts: this one, AS Metadata DPoP Required, and User Authorization Management URI. A newer author who is active in this working group.
Questions for the Working Group
So what should the working group decide at IETF 127?
- Is
client_id_metadata_document_supportedrequired or optional in server metadata? - Is the normative reference to RP Metadata Choices acceptable, and is the plural property the right default?
- Should the
issrequirement and the resource scoping rule be in this document? Both arrived in -03 without a changelog entry. - How does a client learn the selected authentication method when the server publishes no metadata?
- Once those are closed, is one more revision enough to start working group last call?