Practical notes on official Google Chrome, MCP-native actions, persistent sessions, and the failure modes that show up when agents browse real websites.
Forked browsers and hosted automation sessions introduce browser drift and protocol signals. ClawChrome begins from the normal Chrome runtime, then keeps agent control outside the browser.
ClawChrome Team
Product Notes / Guide
Navigate, click, type, observe, and extract are the practical foundation. ClawChrome exposes those actions through MCP-native workflows instead of screenshot-only loops.
Login-gated work breaks when every run starts from a clean browser. Session and cookie persistence give agents a safer path for continuing authenticated workflows.
CDP and WebDriver are useful tools, but protected destinations can observe their control paths. ClawChrome keeps the agent workflow out of that browser-side layer.
Social platforms, account portals, and reputation-sensitive sites fail differently from demo pages. ClawChrome is designed for those high-friction browsing paths.
Provider control planes can add replay, APIs, and cloud packaging, but the failure modes still matter. Compare the runtime, control path, and session behavior first.