[BUG] /login loop or Claude Code, cannot fix on my own

Open 💬 1 comment Opened Jul 1, 2026 by GitScrapMetal

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

/login loop — magic-link flow never completes, no error surfaced, persists for weeks (possible match to backend identity-mapping failure pattern)

What's Wrong?

/login never completes. The flow cycles indefinitely back to the "enter your email" screen with no error thrown, either in the browser or the terminal.

Steps to reproduce:

  1. Run claude in Terminal.app on macOS
  2. Run /login
  3. Select option 1 ("Sign in with your Claude subscription") — options 2 (Anthropic Console/API) and 3 (third-party platform) are not applicable to my use case
  4. Browser opens, I enter my email
  5. I receive the magic-link email and click the link (tested both on the same Mac and on mobile — no difference in outcome)
  6. Browser either returns to the same "enter your email" screen, or shows a code that does not validate when entered in the terminal ("code is not valid" error)
  7. Terminal never completes authentication; no error is thrown — it simply behaves as if /login was never run

This has persisted across multiple weeks and many repeated attempts.

What I've Already Tried (all unsuccessful, symptom identical each time)

  • Removed ANTHROPIC_API_KEY entirely from ~/.zshrc, confirmed unset in a fresh terminal session
  • Updated Claude Code to the latest version via npm install -g @anthropic-ai/claude-code@latest
  • Cleared cached credentials:

``
security delete-generic-password -s "claude-code"
rm -f ~/.claude/.credentials.json
``

  • Manually unlocked Keychain: security unlock-keychain ~/Library/Keychains/login.keychain-db
  • Tried the login flow in an incognito/private browser window
  • Tried completing every step (email open, magic link click, code entry) on the same device (MacBook Air M4) rather than splitting across phone and Mac
  • Confirmed no system-wide ANTHROPIC_API_KEY in /etc/zshrc, /etc/environment, ~/.zprofile, ~/.zshenv, ~/.bash_profile

None of the above changed the outcome. The loop is identical before and after each fix attempt.

Environment

  • Device: MacBook Air M4
  • macOS version: <!-- fill in: Apple menu → About This Mac -->
  • Claude Code version: <!-- fill in: claude --version -->
  • Terminal: Terminal.app (macOS)
  • Auth method attempted: Claude.ai subscription login (option 1 at /login)
  • Subscription tier: <!-- fill in: Pro / Max / Team etc. -->
  • Account status on other surfaces: <!-- fill in: e.g. "Works fine on claude.ai web / mobile app" — this is useful to confirm it's not account suspension -->

Why I suspect this may be related to a known backend issue

This symptom pattern (magic-link/email login not completing, no surfaced error, subscription confirmed active elsewhere) closely matches several previously reported issues describing a backend identity-mapping failure affecting magic-link authentication:

  • #36797 — Authentication redirect loops to onboarding for existing account with active subscription
  • #39788 — OAuth login redirects to account creation instead of authenticating existing Max subscription
  • #51002 — Desktop app connector flow creates duplicate account under same email, locks paid Max subscriber
  • #60022 — Magic-link email login redirects to onboarding for existing Max subscriber (fifth public report of same pattern)
  • #68861 — Magic-link login returns "can't verify email" error on desktop app and web

#60022 specifically notes that Claude Code OAuth login uses the same backend identity-mapping logic as the claude.ai/Desktop magic-link flow, which would explain why this CLI-specific case shows the same behavior.

Support channel history

<!-- Fill in if applicable, e.g.: -->

  • Submitted via support.claude.com on [date] — Conversation ID: [if provided]
  • <!-- Note here if Fin (AI support agent) recommended self-service steps that didn't apply, or did/didn't escalate to a human -->

Impact

This has made Claude Code unusable via subscription login for several weeks. The only workaround is switching to ANTHROPIC_API_KEY, which bypasses subscription billing entirely and incurs pay-as-you-go API charges — not a sustainable substitute for subscription access.

What Should Happen?

/login (option 1) should complete authentication using an active Claude subscription within a single attempt, without requiring device-matched email access, and any failure should surface a clear error rather than silently looping.

Error Messages/Logs

Steps to Reproduce

Steps that are taken to authenticate account from Claude Code to be able to log in.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.197

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗