🐾 claw-stack Back to Plugins
Plugin OpenClaw Ecosystem

Browser Control

Full Chrome DevTools access for your AI agent — automate, debug, and analyze any live web page

Overview

Browser Control bridges your AI agent and a live Chrome instance via the Model Context Protocol. Built on Chrome DevTools and Puppeteer, it exposes 26 tools across 6 capability categories — letting your agent click, type, screenshot, trace performance, inspect network traffic, and execute JavaScript in the real browser, not a headless stub.

Capabilities

Key Features

Reliable UI Automation

Puppeteer-backed click, fill, drag, hover, and keyboard actions with automatic wait-for-result logic — no brittle sleep() hacks.

Accessibility-Tree Snapshots

Capture a full a11y tree of any page in text form, giving the agent stable element UIDs to reference across actions without relying on fragile CSS selectors.

Performance Tracing

Record DevTools traces and extract Core Web Vital scores (LCP, CLS, FID) plus actionable performance insights — with optional CrUX real-user field data.

Network Inspection

List, filter, and inspect every HTTP request and response on the page — including bodies, headers, and error codes — across multiple navigations.

Live Script Execution

Run arbitrary JavaScript functions directly inside the page context and get JSON-serializable results back. Debug state, query the DOM, or trigger custom logic.

Screenshots & Emulation

Capture full-page or element-level screenshots in PNG/JPEG/WebP. Emulate dark mode, device viewports, geolocation, and network throttling conditions.

Architecture

How It Works

The MCP server wraps Chrome's Chrome DevTools Protocol (CDP) and Puppeteer into structured MCP tools. Your agent calls tools over JSON-RPC; the server translates each call into CDP commands over a WebSocket to a live Chrome instance — either one it launches itself or an existing browser you connect via remote debugging port.

MCP Client (Agent) Issues tool calls via JSON-RPC over stdio or SSE
chrome-devtools-mcp Translates MCP tools into Puppeteer / CDP commands
Chrome DevTools Protocol WebSocket channel to the Chrome browser process
Chrome / Chromium Live browser — launched locally or connected remotely
26
MCP Tools
6
Capability Categories
CDP
Protocol Backbone
4
Chrome Channels

Use Cases

What You Can Do

Six capability groups covering the full surface area of a live Chrome browser — from clicking pixels to profiling paint timelines.

UI Automation

  • Click buttons, links, and form controls
  • Type into input fields and text areas
  • Drag-and-drop between page elements
  • Handle browser dialogs and modals
  • Upload files through file inputs
  • Hover and trigger tooltips

Navigation

  • Open and close browser tabs programmatically
  • Navigate by URL, back, forward, or reload
  • Wait for specific text to appear on a page
  • Multi-tab orchestration with page selection
  • Control page loading and cache behavior
  • Manage browser history across sessions

Debugging

  • Capture full-page or element screenshots
  • Take accessibility-tree snapshots of any page
  • Read and filter browser console messages
  • Run arbitrary JavaScript inside the page context
  • Inspect network requests and response bodies
  • Trace performance with Core Web Vitals

Emulation

  • Simulate dark mode or light mode
  • Throttle CPU and network conditions
  • Override geolocation for geo-aware apps
  • Emulate custom viewports and device sizes
  • Spoof user-agent strings
  • Test under Slow 3G / Fast 4G network profiles

Performance

  • Record DevTools performance traces
  • Extract Core Web Vitals (LCP, CLS, FID)
  • Analyze specific performance insights
  • Measure real-user field data via CrUX API
  • Save traces to disk for offline analysis
  • Identify bottlenecks in page load sequences

Network Inspection

  • List all requests since last navigation
  • Filter by resource type (XHR, fetch, media…)
  • Inspect request and response bodies
  • Preserve requests across page navigations
  • Identify failed requests and error codes
  • Export network HAR data for deep analysis

Flexibility

Three Ways to Connect

Whether you want a fresh headless session or to ride along with your existing browser tabs, Browser Control adapts.

Auto-Launch

Default

The MCP server starts Chrome automatically with a dedicated profile. Zero setup required — just configure your MCP client and go.

Uses a persistent user-data-dir so state survives between sessions.

Auto-Connect

Chrome ≥ 144

Attach to your running Chrome instance. Great for mixing manual and agent-driven testing without losing login sessions.

Requires enabling remote debugging in chrome://inspect/#remote-debugging.

Remote Debug Port

Advanced

Point the server at any --remote-debugging-port. Works across VMs, containers, and sandboxed environments over port-forwarding.

Supports custom WebSocket endpoints and Authorization headers.

Coming Soon

Join the Waitlist

Be the first to know when Browser Control launches as an OpenClaw plugin.

Join the Waitlist
Back to Plugins