BrowserClaw

BrowserClaw</h1>
  <p><b>Control your everyday Chrome browser from AI agents, without losing logins or focus.</b></p>
  <p>
    <a href="./docs/MAP.md">Project Map</a> ·
    <a href="./docs/TOOLS.md">Tool Reference (49)</a> ·
    <a href="./AGENT_CONFIG_GUIDE.md">Client Config</a> ·
    <a href="./README.zh-CN.md">Chinese (zh-CN)</a> ·
    <a href="https://github.com/GoldenLoaf24h/browserclaw/releases">Releases</a>
  </p>
</div>

---

<details>
<summary><b>Background: Why BrowserClaw?</b></summary>

<br/>

Browser automation frameworks that drive a separate browser instance (Playwright, Puppeteer, browser-use) start from a clean profile. They do not inherit your active logins, cookies, or extensions, and copying a live Chrome profile on Windows fails with file-sharing locks. Attaching to an existing Chrome via a debug port triggers security banners.

BrowserClaw takes a different route: a Chrome MV3 extension plus a local Native Messaging bridge, running inside the Chrome you already use. Cookies, sessions, and extensions are preserved, and automation happens in background tabs without stealing focus.

</details>

---

## ⚡ What is BrowserClaw?

BrowserClaw is a Chrome extension + local MCP server that lets AI agents operate your real browser. It exposes 49 tools across 7 categories (navigation, perception, action, observation, management, diagnostics, network), with a minimal 14-tool core profile for everyday sessions.

Two execution paths are available:

1. **Deterministic tools** – indexed clicks, fills, batch pipelines, form wizards, screenshots, network capture, etc. The calling agent plans each step.
2. **`chrome_act_toward_goal`** – a local perception-action micro-loop. The native server perceives the page, decides the next action