Source: MCP Client Registration (spec 2026-07-28)

How an MCP client obtains a client_id: pre-registration, Client ID Metadata Documents (preferred for unknown servers) or deprecated Dynamic Client Registration, plus issuer binding.

Contents
  • What: the Client Registration sub-page of the MCP authorization specification
  • Author/publisher: Model Context Protocol project
  • Date: revision 2026-07-28; read 2026-11-14
  • Original: Client Registration

Summary

MCP’s typical case is a client and server with no prior relationship. The page specifies three ways to get a client ID and the order to try them, with Client ID Metadata Documents (CIMD) replacing Dynamic Client Registration (DCR) as the default answer to that case.

Key claims

  • Priority: pre-registered credentials, then CIMD if the AS advertises client_id_metadata_document_supported, then DCR if it has a registration_endpoint, then ask the user.
  • CIMD: the client_id is an HTTPS URL with a path; the JSON document there must contain client_id (exactly equal to the URL), client_name, redirect_uris; clients may authenticate with private_key_jwt.
  • The AS fetches the document, validates it and the redirect URI against it, caches per HTTP headers, and may apply a domain trust policy.
  • DCR (RFC 7591) is deprecated and kept for backward compatibility. Clients must send an application_type (native for desktop/CLI/localhost, web otherwise) because OIDC defaults to web and can reject native redirect URIs.
  • Pre-registered or DCR credentials are bound to the issuing AS’s issuer; if protected-resource metadata points to a new AS, clients must re-register and not reuse them. CIMD IDs are portable across authorization servers.

Feeds

Suggest changes →