How do I use Antigravity Browser Controller?

1. Click on "Deploy Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Antigravity Browser Controller find my open GitHub tab and take a screenshot of it" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

Antigravity Browser Controller

by yihongdo-glitch

Overview Schema Source Related Servers Discussions Admin

Python

Local

Antigravity Browser Controller 🌐⚡ (v2.0.0)

Seamless Native Chrome Automation for AI Agents via Model Context Protocol (MCP)
免重启、免 CDP 端口调试、零幽灵窗口，让 AI 随心接管你日常已登录的 Chrome 浏览器。

Author: yihongdo-glitch License: MIT Protocol: MCP Chrome: Manifest V3 Python: 3.10+ Version: 2.0.0

English | 中文说明

🌐 English Documentation

💡 Why Antigravity Browser Controller v2.0? (The Problem It Solves)

Current AI browser automation approaches—such as Playwright, Selenium, and vision-based OS "Computer Use"—suffer from critical friction points in real-world workflows:

Challenges	Traditional Playwright / CDP	Claude Vision "Computer Use"	Antigravity Browser Controller v2.0
Element Addressing	Fragile CSS selectors / XPath guessing; easily broken by obfuscated class names	Takes heavy screenshot & guesses coordinates; burns high-cost Vision tokens	Ref Snapshot (`page.snapshot`): Numbered refs (`[12] button 'Submit'`), 85% token reduction, zero selector brittleness
Login Persistence (Cookies)	Requires separate profile; loses sessions; triggers aggressive captchas/bot blocks	Operates in active desktop foreground; disrupted by user input	100% inherits your existing Chrome logins (no re-auth, bypasses captchas)
Multi-Agent Concurrency	Single debugger port locks out secondary agents	Fullscreen lock; cannot share mouse	Dual-port decoupled architecture (WS:18888 + HTTP:18889): concurrent, stateless, zero-lockout
Background & Lock Screen	Creates blank `about:blank` ghost windows that hijack focus	Cannot run in background; fails immediately when screen is locked (`Win+L`)	Runs silently in memory & DOM; continues operating even when Windows is locked
Modern Web Frameworks	Form fills fail to trigger React / Vue synthetic state changes	Coordinate clicks are prone to 5px offsets on dynamic or scrolling pages	Overrides native setters to guarantee 100% reliable React / Vue event dispatching
Security & CSRF	Unauthenticated local ports expose browser to malicious websites	N/A	Origin isolation + Localhost Bearer Token auth (`~/.antigravity/bridge.token`)

🚀 10-Second Quickstart

Step 1: Install the Chrome Extension

git clone https://github.com/yihongdo-glitch/antigravity-browser-mcp.git

Open Chrome and navigate to chrome://extensions/.
Toggle Developer mode in the top-right corner to ON.
Drag and drop the cloned project folder directly into the extensions page (or click Load unpacked and select the folder).
The purple Antigravity icon will appear in your Chrome toolbar.

Step 2: Configure Your AI Client (One-Click MCP Integration)

1. Claude Desktop (`claude_desktop_config.json`)

Add the following to your mcpServers object:

{
  "mcpServers": {
    "antigravity-browser": {
      "command": "python",
      "args": [
        "path/to/antigravity-browser-mcp/bridge/mcp_server.py"
      ]
    }
  }
}

2. Cursor / Windsurf

In Settings -> Features -> MCP, add a new stdio server:

Name: antigravity-browser
Type: command
Command: python path/to/antigravity-browser-mcp/bridge/mcp_server.py

3. Antigravity IDE

Pre-registered in mcp_config.json, fully plug-and-play.

🛠️ MCP Tools Reference (Full Capabilities)

Tool Name	Key Parameters	Description
`browser_page_snapshot`	`tab_id`, `max_elems`	AI Page Eyes: Analyzes visible interactive elements, assigning unique numbered refs (`[1]`, `[2]`). Use before click/fill!
`browser_page_content`	`tab_id`, `max_len`	Smart Content Extraction: Cleanly extracts main article body text while discarding noise.
`browser_page_wait`	`selector`, `text`, `timeout`	Waits until a specific CSS selector or visible text snippet appears on the page.
`browser_click_element`	`ref`, `selector`, `text`	Clicks an element. Prioritizes `ref`, followed by deepest leaf-node text match and interactive ancestor escalation.
`browser_fill_input`	`value`, `ref`, `selector`	Injects text into inputs/textareas while bypassing React/Vue synthetic DOM barriers.
`browser_press_key`	`key`, `ref`, `selector`	Simulates keyboard key presses (`Enter`, `Tab`, `Escape`, `ArrowDown`, etc.). Submits forms automatically on Enter.
`browser_scroll_page`	`ref`, `dx`, `dy`, `to`	Scrolls the page or scrolls a specific element ref into view (`to="top"` or `to="bottom"`).
`browser_list_tabs`	none	Lists all open tabs in your Chrome window (ID, title, URL, active state).
`browser_get_active_tab`	none	Retrieves metadata of the currently focused browser tab.
`browser_find_tab`	`url_keyword`, `title_keyword`	Instantly finds an open tab matching a URL or title substring (e.g., `'zhipin.com'`, `'github'`).
`browser_evaluate_script`	`tab_id`, `script`	Executes arbitrary JavaScript in the page context, automatically bypassing page CSP restrictions.
`browser_take_screenshot`	`tab_id`, `output_path`	Captures high-fidelity page screenshots—works even when Chrome is minimized or screen is locked.
`browser_send_cdp`	`tab_id`, `method`, `params`	Direct Chrome DevTools Protocol escape hatch.
`browser_navigate`	`tab_id`, `url`	Navigates the specified tab to a new URL.
`browser_create_tab`	`url`, `active`	Opens a new browser tab.
`browser_close_tab`	`tab_id`	Closes a specific tab by ID.

🏗️ Architecture & Privacy Guarantee

  [ AI Clients: Claude Desktop / Cursor / Antigravity / CLI Scripts ]
                         ▲
                         │ (HTTP POST REST Gateway: http://127.0.0.1:18889/action)
                         │ (Authenticated via ~/.antigravity/bridge.token)
                         ▼
        [ Antigravity Bridge Daemon & WebSocket Hub ]
                         ▲
                         │ (Local WebSocket: ws://127.0.0.1:18888/ws)
                         │ (Origin-filtered: chrome-extension:// ONLY)
                         ▼
   [ Chrome Extension v2.0 (Manifest V3 Service Worker) ]
                         │
        ┌────────────────┴────────────────┐
        ▼                                 ▼
[ Isolated World DOM Sandbox ]    [ Chrome DevTools Protocol (CDP) ]
- window.__agRefs (Ref Model)     - CSP Bypass Runtime.evaluate
- Leaf-Node Precision Engine      - Lock-Screen Silent Screenshot
- Controlled Component Bypass

🔒 Privacy Commitment: All communication happens strictly over the local loopback interface (127.0.0.1) with origin filtering and token authentication. No external servers, no cloud relays, zero telemetry, zero analytics. Your account cookies, session tokens, and browsing data never leave your local machine.

Related MCP server: sansan-chrome-mcp

🇨🇳 中文说明文档

💡 为什么选择 Antigravity Browser Controller v2.0？(痛点对比)

目前主流的 AI 浏览器控制方案（Playwright / Selenium / 视觉截图 OS Control）普遍存在以下痛点：

痛点维度	传统 Playwright / CDP 方案	Claude 视觉“操作电脑”	Antigravity Browser Controller v2.0
元素定位与寻址	脆弱的 CSS 选择器/XPath，页面动态哈希 class 变动即失效	截全屏估算物理坐标，极耗高额视觉 Token	Ref 编号快照 (`page.snapshot`)：将可交互元素直接编号为 `[12] button '立即沟通'`，Token 暴降 85%，零盲猜
日常登录态 (Cookie)	必须使用独立用户目录，掉登录态，极易触发滑块风控	必须在前台全屏操作，易受外界干扰	100% 继承你日常 Chrome 中所有网站的已登录状态
多 Agent 并发与锁死	调试端口互斥，二次拉起直接端口冲突崩溃	独占物理桌面，无法并发	双端口解耦 (WS:18888 + HTTP:18889)：无状态 HTTP 控制网关，多 Agent/脚本并发零锁死
后台与锁屏	经常弹死白空白窗口（`about:blank`）遮挡屏幕	无法在后台运行；锁屏 (`Win+L`) 即报错崩溃	纯静默后台与内存 DOM 操作，锁屏后全自动运行
前端框架兼容性	注入表单常因 React/Vue 受控组件虚拟 DOM 而无法提交	依靠物理坐标点击易偏 5 像素	重写 Native Setter，穿透并触发真实变更事件
本地网络安全	本地无鉴权端口，易受网页 CSRF / DNS Rebinding 攻击	无	严格 Origin 校验 + 本地 Bearer Token 随机密钥 (`~/.antigravity/bridge.token`)

🚀 10 秒极速上手

步骤 1：安装 Chrome 扩展

git clone https://github.com/yihongdo-glitch/antigravity-browser-mcp.git

打开 Chrome 地址栏输入并回车：chrome://extensions/
打开右上角 「开发者模式」 开关。
将本项目文件夹直接拖入该页面（或点击左上角「加载已解压的扩展程序」选择本项目目录）。
工具栏出现紫色 Antigravity 图标即表示就绪！

步骤 2：配置你的 AI 客户端 (MCP 一键挂载)

1. 在 Claude Desktop 中使用

在 claude_desktop_config.json 的 mcpServers 节点中添加：

{
  "mcpServers": {
    "antigravity-browser": {
      "command": "python",
      "args": [
        "path/to/antigravity-browser-mcp/bridge/mcp_server.py"
      ]
    }
  }
}

2. 在 Cursor / Windsurf 中使用

在 Settings -> Features -> MCP 中添加自定义 stdio 命令：

Name: antigravity-browser
Command: python
Args: path/to/antigravity-browser-mcp/bridge/mcp_server.py

3. 在 Antigravity 中使用

已内置挂载至全局 mcp_config.json，开箱即用。

🛠️ MCP 核心工具能力一览

工具名 (Tool)	关键参数 (Params)	描述 (Description)
`browser_page_snapshot`	`tab_id`, `max_elems`	AI 页面之眼：智能分析当前页面的所有可见交互元素，分配唯一编号 Ref（如 `[1]`, `[2]`...），点击/填写前必调！
`browser_page_content`	`tab_id`, `max_len`	智能正文抽取：自动识别文章/主内容容器，剥离无关导航和噪音。
`browser_page_wait`	`selector`, `text`, `timeout`	异步等待指定选择器或文字在页面中渲染完成。
`browser_click_element`	`ref`, `selector`, `text`	精准点击。优先支持 Ref 编号，文本模式具备最深叶子节点匹配 + 向上交互语义穿透，杜绝点偏死容器。
`browser_fill_input`	`value`, `ref`, `selector`	自动输入文本，支持 Ref 编号，穿透 React/Vue 虚拟 DOM，100% 触发 input/change 事件。
`browser_press_key`	`key`, `ref`, `selector`	模拟按键（`Enter`, `Tab`, `Escape`, 箭头按键等）。在 input 上按 Enter 自动触发表单提交。
`browser_scroll_page`	`ref`, `dx`, `dy`, `to`	滚动页面或将指定 Ref 元素居中对齐，支持 `to="top"` / `to="bottom"`。
`browser_list_tabs`	无	列出当前 Chrome 中打开的所有标签页（ID、标题、网址、激活状态）。
`browser_get_active_tab`	无	获取当前正在浏览的活动标签页详情。
`browser_find_tab`	`url_keyword`, `title_keyword`	按网址或标题关键词瞬间找到指定标签页（如搜索 `"zhipin.com"`）。
`browser_evaluate_script`	`tab_id`, `script`	在页面上下文执行任意 JS 脚本，底层自动绕过 CSP 安全限制。
`browser_take_screenshot`	`tab_id`, `output_path`	高清截图，即使浏览器在后台或电脑锁屏 (`Win+L`) 也能清晰截取。
`browser_send_cdp`	`tab_id`, `method`, `params`	直接发送 Chrome DevTools Protocol 指令（逃生舱）。
`browser_navigate`	`tab_id`, `url`	控制标签页跳转到新网址。
`browser_create_tab`	`url`, `active`	新建标签页。
`browser_close_tab`	`tab_id`	关闭指定标签页。

🏗️ 架构与安全设计

  [ AI 客户端 (Claude / Cursor / Antigravity / 脚本) ]
                         ▲
                         │ (无状态 HTTP REST 网关: http://127.0.0.1:18889/action)
                         │ (Bearer Token 鉴权: ~/.antigravity/bridge.token)
                         ▼
        [ Python 桥服务守护进程 (WebSocket + HTTP) ]
                         ▲
                         │ (ws://127.0.0.1:18888/ws 本地环回)
                         │ (Origin 过滤: 仅放行 chrome-extension://)
                         ▼
      [ Chrome 扩展 v2.0 (Manifest V3 Service Worker) ]
                         │
        ┌────────────────┴────────────────┐
        ▼                                 ▼
[ 页面 Isolated DOM 沙箱 ]          [ CDP 调试桥接 ]
- window.__agRefs (Ref 快照模型)    - 绕过 CSP 注入
- 叶子节点精准穿透引擎              - 锁屏后台静默快照
- React/Vue 受控组件穿透

🔒 隐私与安全承诺：本工具完全基于本地环回地址（127.0.0.1）进行进程间通讯，具备 Origin 来源过滤与本地 Bearer Token 鉴权机制。不存在任何外部服务器、没有任何云端中继、零隐私数据上报。你的所有账号 Cookie 和浏览数据绝不离开你的电脑本地。

📄 License

This project is licensed under the MIT License - created by @yihongdo-glitch. Free for personal and commercial use.

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained

ResponsivenessNo issues

Resources

GitHub Repository

Need Help?

Related Servers

Related MCP Connectors

agent-web-actions
com.trycloudflare.clients-involving-billion-headers
Stealth web automation for AI agents. Login, signup, navigate, screenshot.
Browser Automation Web Scraping
2
agent-web-actions
com.trycloudflare.plates-spread-materials-optimum
Stealth web automation for AI agents. Login, signup, navigate, screenshot.
Browser Automation Agent Orchestration
2
TinyFish AI Web AgentOAuth
ai.tinyfish.agent
AI-powered web automation. Navigate websites using AI agents for one page or a thousand
Browser Automation Agent Orchestration Web Scraping
2
TinyFish AI Web AgentOAuth
ai.mino
AI-powered web automation. Navigate websites using AI agents for one page or a thousand
Browser Automation Autonomous Agents Web Scraping
2

Related MCP Servers

agent-browser-mcp
Browser Automation Web Scraping OS Automation
335234131
A
license
B
quality
F
maintenance
Enables AI agents to directly control your real Chrome browser with full context including login sessions, cookies, and open tabs. It provides tools for page scanning, JavaScript execution, CDP control, screenshots, and physical mouse/keyboard input for authentic browser automation.
Updated 5 months ago (2026-04-15 12:32 UTC)
20
243
MIT
sansan-chrome-mcp
Browser Automation Web Scraping
AbellLee
A
license
Not graded
quality
A
maintenance
Enables AI assistants to control Chrome browser operations such as navigation, reading pages, taking screenshots, managing tabs, and more via a Chrome extension and native messaging.
Updated 2 months ago (2026-08-05 03:03 UTC)
MIT
Invictum Browser Bridgeofficial
Browser Automation Agent Orchestration
invictumhr
A
license
Not graded
quality
B
maintenance
Enables AI agents to securely control a user's existing Chrome profile locally, providing typed browser actions, form and editor support, WordPress workflows, terminal automation, and Figma inspection with policy-based authorization and redacted auditing.
Updated 9 days ago (2026-09-14 14:18 UTC)
MIT
browser-control
Browser Automation Agent Orchestration
kierandrewett
F
license
Not graded
quality
A
maintenance
Enables AI agents to control existing Chrome and Firefox browser sessions locally, including tab management, navigation, form filling, screenshots, and persistent scripts, without remote relays or debugging ports.
Updated 18 days ago (2026-09-05 21:11 UTC)
9 npm
-