Official Google Chrome
ClawChrome runs real google-chrome-stable instead of a Chromium fork, patched browser, or hosted browser abstraction. Agents start from the same browser baseline a person would use.
Features
Browser control built for agents, not test scripts.
ClawChrome gives agents fast isolated Chrome sessions, direct MCP/API control, useful page-state output, session viewing, and persistent browser context for workflows that break under raw CDP, WebDriver, or hosted browser sessions.
observe current page
Cookie preferences
Returned to the agent
visible_layer: cookie_preferences
clickable: Accept, Customize
blocked: page buttons behind overlay
The system calculates the page layers structurally. The agent receives the state it can act on, instead of a misleading list of elements hidden behind a blocking overlay.
Problem
Most browser tools were not designed for autonomous use.
- CDP and Playwright are excellent testing and debugging tools, but brittle when an agent has to drive unknown production websites.
- Banners, modals, native prompts, and overlays often hide the page from the agent or make the next action ambiguous.
- Proxy flags, patched fingerprints, and forked browser runtimes create browser profiles that do not match a normal desktop Chrome session.
Core features
The runtime pieces that make agent browsing less brittle.
Lightweight isolated sessions
Each agent workflow can get a separate Chrome session instead of taking over a user browser or running inside a heavyweight shared desktop. Sessions can start, idle, resume, and persist around the workflow.
Direct MCP/API control
Agents can navigate, observe, click, type, scroll, and extract through MCP-native actions or direct product API control. CDP and WebDriver do not have to be the center of the browser workflow.
Custom OS browser control
Actions are translated outside browser-side automation protocols that sites commonly inspect. Agents get practical browser control while Chrome stays closer to a normal desktop session.
Structural page-layer handling
ClawChrome calculates the page layers so modal overlays, cookie banners, and blocking elements change what the agent sees and can click, the same way they do for a human user.
Agent-focused snapshots
The browser output is scoped to the useful interactive state of the page, not a raw debugging dump. When an element is blocked, the agent can be told what blocks it instead of guessing.
Persistent sessions and cookies
Chrome sessions can keep cookies and browser state across agent runs. When account access is needed, cookie sync can move an authenticated session into the agent workflow.
Session viewing
Teams can inspect the browser session while the agent works through direct low-latency streaming, instead of relying only on screenshots, logs, or replay after a run has already failed.
Custom OS network routing
Proxy configuration lives below Chrome instead of inside Chrome flags. Browser traffic is routed through the custom OS layer so the browser profile stays consistent with the network path.
01
Start a session
Create a lightweight isolated Chrome session for the agent instead of attaching to a user browser or launching a fork.
02
Connect control
Give the agent MCP-native or direct API browser actions for navigation, observation, clicking, typing, and extraction.
03
Use the page state
Return the elements that are actually available through the current page layers, including modal and banner state.
04
Stream or persist
View the browser session while the agent works, then idle, resume, or persist cookies and session data for repeat workflows.
FAQ
Short answers for evaluators.
Is ClawChrome a Chrome fork?
No. ClawChrome is built around official Google Chrome running in a lightweight isolated session with a custom OS control layer around it.
Is this a heavy virtual desktop product?
No. The product direction is fast isolated Chrome sessions for agents, not a heavyweight shared desktop. The exact packaging can vary by deployment, but the operator experience should feel like starting, controlling, viewing, idling, and resuming browser sessions.
Does ClawChrome replace Playwright?
For agent browsing workflows, yes. Playwright and CDP are still useful for testing, but ClawChrome is designed for agents driving real sites where page state, overlays, sessions, and browser profile consistency matter.
Why does page-layer handling matter?
Agents need to know what can actually be interacted with. If a cookie banner blocks a button, returning the underlying button as clickable makes the workflow brittle. ClawChrome scopes the output to the current interactive state.
Can I use my own proxy?
Yes. The important difference is that proxy routing is handled below Chrome, rather than only by setting Chrome proxy flags.
Start with the setup path.
Connect an MCP client, start a Chrome session, and test the workflow that currently breaks under raw browser automation.