[BUG] Claude Code Web — Session resume after Chrome restart triggers 502/CSP/JSON cascade errors, session stuck indefinitely

Resolved 💬 3 comments Opened Mar 24, 2026 by julienmerconsulting Closed Mar 28, 2026

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?

Resuming a Claude Code web session (claude.ai/code) after a full Chrome restart consistently triggers a cascade of HTTP errors, CSP violations, and JSON parse failures. The session becomes unusable — stuck on "Clauding..." spinner indefinitely. Recurring issue, reproducible every time Chrome is fully closed and relaunched.

What Should Happen?

The session should resume cleanly after a browser restart, re-establishing all connections without errors and restoring the working state correctly.

Error Messages/Logs

# HTTP errors
GET /api/organizations/3.aude-opus-4-6[...] → 404 Not Found
  → Malformed URL generated by the UI itself (invalid model string in path)

GET /v1/session ingress/..git_proxy/compare → 502 Bad Gateway (multiple occurrences)
  → Gateway failure on session resume, not transient — persists across page reloads

# Content Security Policy violations
Fetch API cannot load https://a-api.anthropic.com/v1/m
  → Refused to connect: violates Content Security Policy directive "connect-src 'self'"
  → The UI is generating requests to a non-whitelisted endpoint

Error sending segment performance metrics TypeError: Failed to fetch
  → Refused to connect: violates Content Security Policy (remote-metrics.ts:53)

# JSON parse errors (multiple)
[REACT_QUERY_CLIENT] QueryClient error: SyntaxError: 
  Unexpected token '<', "<!DOCTYPE..." is not valid JSON
  → Server returning HTML error page instead of JSON — likely 502/504 upstream response 
     being passed through as-is to the React Query client

# MCP
[MCP] Pretending GDrive is MCP: gcal_and_gmail_mcp is enabled but mcp_gdrive is not
  → MCP state inconsistency on session restore

Steps to Reproduce

  1. Open claude.ai/code in Chrome
  2. Start or resume a Claude Code session and work actively (mid-task)
  3. Close Chrome entirely (not just the tab)
  4. Relaunch Chrome
  5. Navigate back to the exact session URL (claude.ai/code/session_XXXX)
  6. Open DevTools → Console tab
  7. Observe: cascade of 502, 404, CSP errors fires immediately on page load
  8. Session UI stuck on "Clauding..." spinner — no recovery, no timeout message

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

N/A — Web version (claude.ai/code)

Platform

Other

Operating System

Windows

Terminal/Shell

Other

Additional Information

  • Hard refresh (Ctrl+Shift+R) does not resolve the issue
  • Opening the session URL in a new tab does not resolve the issue
  • The 404 on /api/organizations/3.aude-opus-4-6 suggests the UI is constructing

a malformed API path — possibly a model identifier being incorrectly interpolated
into an organization endpoint URL

  • The CSP violations indicate the UI is attempting to reach endpoints not listed

in the Content Security Policy, suggesting a mismatch between the deployed
CSP config and the actual endpoints the frontend tries to call

  • Screenshots attached (console errors visible)

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗