Agents repeatedly stuck at login for browser UI validation (claude-in-chrome 'not connected' + no sign-in path)

Summary

Across multiple Claude Code sessions (including parallel ones), the agent is repeatedly stuck at login when it tries to do basic browser-based UI testing / validation. Two compounding problems:

1. The Claude for Chrome (claude-in-chrome) extension keeps dropping its connection. mcp__claude-in-chrome__* tools intermittently fail with "Browser extension is not connected" mid-session, even with Chrome open and logged into claude.ai. Running /chrome reconnects it, but it recurs (and recurs across separate sessions).
2. Once connected, there is no way for the agent to get past an app login wall. Navigating to an authenticated web app redirects to its login page, and the agent has no sanctioned way to authenticate (it can't enter credentials), so it stalls until a human logs in manually.

Net effect: the agent cannot autonomously verify its own UI work in a browser. The core agentic-dev loop — make a change → validate it in a browser → ship — breaks in essentially every session, and a human has to step in just to get past login.

Environment

• Claude Code: 2.1.185
• OS: macOS 26.5.1 (arm64)
• Chrome: stable channel (exact build not captured)

Verbatim error (returned by tabs_context_mcp)

Browser extension is not connected. Please ensure the Claude browser extension is installed and running (https://claude.ai/chrome), and that you are logged into claude.ai with the same account as Claude Code. If this is your first time connecting to Chrome, you may need to restart Chrome for the installation to take effect.

Repro

1. Run a Claude Code session (or several in parallel) that use mcp__claude-in-chrome__* to validate a UI change.
2. Call tabs_context_mcp → intermittently returns "not connected" despite Chrome being open and logged into claude.ai.
3. /chrome reconnects it temporarily; the disconnect recurs later and in other concurrent sessions.
4. When the extension is connected, navigate to an authenticated web app → redirected to its login page → the agent cannot proceed → validation stalls until a human manually logs in.

Impact

Browser-based UI verification and testing cannot be run autonomously. This blocks validation across all of my sessions and undercuts the main value of agentic coding (the agent verifying its own work before shipping).

Asks

1. Make the extension connection reliable / auto-recovering so it does not drop mid-session and does not require a manual /chrome per session.
2. Provide a first-class, secure way for an agent to authenticate to a web app for validation — e.g. a persistent/shared browser session, or a sanctioned credential handoff for sign-in — so agents are not permanently stuck at the login wall.