Under the hoodTier: how much it should matter to you
2Useful: my rating, 1 to 5
2Signal: worth watching, 1 to 5
SessionsArea: what it touches
Internal ChangesKind: in v2.1.295,
Internal ChangesSection of the release
What
Claude Code can load a configuration document from a URL given with --client-data-url or the CLAUDE_CODE_CLIENT_DATA_URL environment variable (a setting passed in through your shell). This release changes when that document is loaded and what it can do:
Loading now starts early in startup (startClientDataUrlFetch), before the main load. The later load reuses this early copy if the URL has not changed, and it re-checks the allow_model_catalog policy.
A new clientDataUrl option feeds the early load, and how long it takes is now measured.
The early load is skipped for third-party providers, remote sessions, host-managed providers, an invalid URL, and when policy helpers are set.
An entry in the document with assume_first_party_base_url can now make the base URL (the address Claude Code sends its API requests to) be treated as Anthropic's own API. That setting is withdrawn if the launch does not use the document it came from.
Two internal checks of _CLAUDE_CODE_ASSUME_FIRST_PARTY_BASE_URL now go through helper functions instead of reading the environment variable directly.
For remote workers, startup also finds a fallback toolContainerToken through a new helper, so the agent proxy gets a token even when none is passed in.
Why
Starting the load early can shorten startup when you use a client-data URL. If you point Claude Code at a custom base URL, a client-data document can now decide whether that URL is treated as Anthropic's own API, which can change any behaviour that depends on talking to Anthropic directly.
