[Bug] --chrome flag ineffective when resuming session — Chrome MCP stays disabled
Description
When resuming a session with claude --resume --chrome, the Chrome MCP server is not started. /chrome shows Status: Disabled. The --chrome flag is silently ignored on resume.
This is increasingly problematic as long-running sessions become the norm — users shouldn't have to abandon their session context just to get browser access back.
Steps to Reproduce
- Start a session:
claude --chrome(Chrome MCP works) - Exit the session
- Resume:
claude --resume --chrome - Run
/chrome→ Status: Disabled - Chrome MCP tools (
mcp__claude-in-chrome__*) are unavailable
Expected Behavior
--chrome on resume should spawn a fresh Chrome MCP server, just like it does on a new session.
Actual Behavior
The --chrome flag is ignored. No chrome-native-host process is spawned. No bridge sockets are created. The MCP server is never started.
Environment
- OS: Ubuntu Linux (kernel 6.17)
- Claude Code: Latest (March 2026)
- Chrome: Installed with Claude extension and native messaging host configured
- Session type: Long-running (days to weeks, surviving multiple compactions)
Context
This is distinct from #25956 (WebSocket bridge death in long-lived sessions). That bug is about an active Chrome MCP connection dying mid-session. This bug is about the connection never being established on resume.
The combination makes Chrome MCP particularly fragile for long-running sessions:
- If Chrome MCP dies mid-session (#25956) → can't recover without restart
- If you resume with
--chrome→ flag is ignored, MCP never starts - Only workaround: completely new session (
claude --chromewithout--resume)
This forces users to choose between session continuity and browser access, which shouldn't be necessary.
Workaround
Start a completely new session (not resume) with claude --chrome. This loses the conversation context from the previous session.
This issue has 5 comments on GitHub. Read the full discussion on GitHub ↗