If you have ever tried to run an AI agent in Chrome while working, you know the pain of the browser stealing focus every time it executes a command. A new tool called agent-chrome, posted to Hacker News on October 2, 2026, solves this specific macOS headache for Claude Code users. Developed by rav4nn, this open-source solution allows Claude to browse using a copy of your signed-in Chrome profile in a background window, ensuring your keyboard and screen real estate remain under your control.

The Focus Theft Problem

The core issue stems from how Chrome handles remote debugging on macOS. When tools like Playwright MCP or Chrome DevTools MCP interact with the browser via the debug port, Chrome aggressively brings the window to the foreground. This interrupts developer workflow, forcing constant context switching. Headless browsers avoid this focus theft but hide the agent's actions, making it impossible to intervene during logins, 2FA prompts, or CAPTCHAs. agent-chrome bridges this gap by providing a visible, interactive window that stays behind your terminal.

Proxy Architecture and Multiplexing

Under the hood, agent-chrome runs a lightweight Node.js proxy that holds a --remote-debugging-pipe connection to Chrome. This pipe is exclusive to the launching process, preventing other clients from connecting directly. The proxy exposes a localhost HTTP and WebSocket interface compatible with standard Chrome DevTools Protocol discovery. It handles multiplexing by remapping request IDs from multiple Claude sessions to unique proxy IDs and caching target states. This allows several Claude Code sessions to share a single signed-in Chrome instance without fighting over profile locks or stealing focus.

Security and Profile Handling

The tool does not use your default Chrome profile directly because Chrome refuses remote debugging on default data folders. Instead, it creates an APFS clone of your profile, which consumes no extra disk space until changes diverge. For security, the proxy strictly rejects requests with an Origin header or non-localhost Host headers to prevent web-based attacks. It also disables extensions in the agent window to maintain a clean state, though this means your existing password managers won't work inside the red-themed agent window.

Key Takeaways

  • agent-chrome enables background browsing for Claude Code on macOS without stealing focus.
  • It uses a Node.js proxy to multiplex multiple Claude sessions over a single Chrome pipe.
  • The tool creates an APFS clone of your Chrome profile to bypass default directory restrictions.
  • It is currently macOS-only, free, open-source, and licensed under MIT.

The Bottom Line

This is a clever hack that respects the developer's need for visual oversight without sacrificing workflow momentum. By decoupling the debug port from the focus state, agent-chrome makes agentic browsing actually usable in a real dev environment.

Technical Implementation Details

The proxy is a fork of mimkorn/chrome-pipe-proxy with significant additions for Claude Code integration. It includes a setup CLI and a Claude Code plugin that teaches the LLM how to interact with the tool. The proxy launches Chrome only when a client opens a tab, and it handles Page.bringToFront internally to prevent the window from raising itself. This architectural choice ensures that while the agent is visible, it remains passive to the user's active window management.