Cookie domain filtering was wrong — the session token domain is unknown
and was never found by claude.ai/anthropic.com filters.
LoginView: replace getAllCookies domain check with callAsyncJavaScript
that fetches /api/bootstrap directly from the WebView. The WebView uses
its own full session (all cookies, any domain) so auth is detected
correctly regardless of where the token lives.
UsageViewModel: claudeCookieHeader now sends all cookies from the app's
WKWebsiteDataStore instead of filtering by domain. checkInitialSignInState
likewise checks for any cookie.
Claude session cookies are on anthropic.com, not claude.ai. The login
window was not detecting auth (Cancel stayed, no Done) and API calls
were sent without the actual session token.
- claudeCookieHeader: include anthropic.com cookies so the token is
sent to the usage/bootstrap endpoints
- checkInitialSignInState: detect anthropic.com cookies on startup
- didFinish in LoginView: fire auth when anthropic.com cookies found
- signOut: clear anthropic.com data alongside claude.ai
- notAuthenticated catch: set isSignedIn = false so Settings stays
in sync with the main panel
Claude's usage/prepaid/overage endpoints require Origin, Referer, and
User-Agent headers to pass CORS/auth checks. Without them, bootstrap
succeeds (more permissive) but usage returns 403 -> 'Not signed in'.
Added claudeAPIRequest(for:) helper that sets all required browser-like
headers on every request. Bootstrap, usage, prepaid, overage, and the
orgs fallback all go through it.
Loading https://claude.ai as the start URL caused didFinish to fire on
the landing page while stale/tracking cookies were already in
WKWebsiteDataStore. The 'any claude.ai cookie' check then fired
immediately, closing the login sheet before the user could sign in.
Loading /login ensures the URL-guard catches the initial page load and
only checks cookies after the real post-login redirect.
github-actions[bot]
superdooper86
SuperDooper