← 返回 Skills 市场
femto

Browser Automation

作者 femto · GitHub ↗ · v0.1.0
cross-platform ⚠ suspicious
709
总下载
2
收藏
3
当前安装
1
版本数
在 OpenClaw 中安装
/install browser-automation-2
功能描述
Control Chrome browser with AI using MCP protocol. Use when users want to automate browser tasks, take screenshots, fill forms, click elements, navigate page...
使用说明 (SKILL.md)

Browser Automation (Chrome MCP Server)

Turn your Chrome browser into your intelligent assistant - Let AI take control of your browser.

When to Use This Skill

Use this skill when the user:

  • Wants to automate browser tasks (clicking, filling forms, navigating)
  • Needs to take screenshots of web pages or elements
  • Wants to extract content from web pages
  • Asks to search browsing history or manage bookmarks
  • Needs to monitor network requests
  • Wants AI to interact with websites using their existing login sessions

Installation

Step 1: Install the Native Bridge

npm install -g mcp-chrome-bridger
# or
pnpm install -g mcp-chrome-bridger
mcp-chrome-bridger register

Step 2: Install Chrome Extension

Download from GitHub Releases:

  1. Download mcp-chrome-extension-vX.X.X.zip
  2. Open Chrome → chrome://extensions/
  3. Enable "Developer mode"
  4. Click "Load unpacked" and select the extracted folder
  5. Click the extension icon → Click "Connect"

Step 3: Configure MCP Client

Add to your MCP client configuration:

Streamable HTTP (Recommended):

{
  "mcpServers": {
    "chrome-mcp-server": {
      "type": "http",
      "url": "http://127.0.0.1:12306/mcp"
    }
  }
}

STDIO (Alternative):

{
  "mcpServers": {
    "chrome-mcp-server": {
      "command": "npx",
      "args": ["mcp-chrome-bridger", "stdio"]
    }
  }
}

Available Tools (20+)

Browser Management

Tool Description
get_windows_and_tabs List all browser windows and tabs
chrome_navigate Navigate to URLs, control viewport
chrome_switch_tab Switch active tab
chrome_close_tabs Close specific tabs
chrome_go_back_or_forward Browser history navigation

Screenshots

Tool Description
chrome_screenshot Capture full page, viewport, or specific elements

Content Analysis

Tool Description
chrome_get_web_content Extract HTML/text from pages
chrome_get_interactive_elements Find clickable elements
search_tabs_content AI-powered semantic search across tabs
chrome_console Capture browser console output

Interaction

Tool Description
chrome_click_element Click elements via CSS selector
chrome_fill_or_select Fill forms and select options
chrome_keyboard Simulate keyboard input

Data Management

Tool Description
chrome_history Search browsing history
chrome_bookmark_search Find bookmarks
chrome_bookmark_add Add new bookmarks
chrome_bookmark_delete Delete bookmarks

Network

Tool Description
chrome_network_capture_start/stop Monitor network requests
chrome_network_request Send HTTP requests with browser cookies

Example Usage

Navigate and Screenshot

User: "Take a screenshot of github.com"

AI uses:
1. chrome_navigate(url: "https://github.com")
2. chrome_screenshot(fullPage: true)

Fill a Form

User: "Login to my account on example.com"

AI uses:
1. chrome_navigate(url: "https://example.com/login")
2. chrome_fill_or_select(selector: "#email", value: "[email protected]")
3. chrome_fill_or_select(selector: "#password", value: "...")
4. chrome_click_element(selector: "button[type=submit]")

Search History

User: "Find all pages I visited about React hooks last week"

AI uses:
1. chrome_history(text: "React hooks", startTime: "1 week ago")

Extract Content

User: "What does this page say about pricing?"

AI uses:
1. chrome_get_web_content()
2. Analyzes the extracted content

Advantages Over Playwright

Feature Playwright MCP Chrome MCP Server
Browser Instance New browser process Your existing Chrome
Login Sessions Need to re-login Uses existing sessions
User Settings Clean environment Your bookmarks, extensions, settings
Startup Time Slow (launch browser) Instant (extension already loaded)
Resource Usage Heavy Lightweight

Multi-Client Support

Multiple AI clients can connect simultaneously:

  • Claude Code
  • Cursor
  • Kiro
  • Any MCP-compatible client

Each client gets its own session while sharing the same Chrome browser.

Troubleshooting

Extension Not Connecting

  1. Check extension is enabled in chrome://extensions/
  2. Click extension icon → Verify "Connected" status
  3. Restart Chrome if needed

Port Already in Use

The server automatically handles port conflicts. If issues persist:

lsof -i :12306
kill \x3CPID>

Resources

安全使用建议
This skill appears to do what it says (control your installed Chrome via an MCP bridge and extension), but before installing anything you should: 1) verify the npm package (mcp-chrome-bridger) and the GitHub project (mcp-chrome) are the legitimate projects you expect (check publisher, repository activity, code, issues, and release artifacts); 2) understand that the extension + bridge will be able to read and act on your cookies, logged-in sessions, browsing history, bookmarks, console output, and network traffic—only install if you trust the project and need that level of access; 3) prefer installing and testing on a disposable or secondary Chrome profile (not your main profile) to limit exposure; 4) require explicit user consent before the agent performs sensitive actions (logins, form submission, network captures); 5) after testing, remove/uninstall the extension and npm package if you have doubts and rotate any accounts that may have been accessed. If you want more certainty, ask the skill provider for the exact repository, package publisher details, and signatures or review the extension and package source code yourself.
功能分析
Type: OpenClaw Skill Name: browser-automation-2 Version: 0.1.0 The skill is classified as suspicious due to the inclusion of the `chrome_network_request` tool, which explicitly allows sending HTTP requests 'with browser cookies'. While presented as a legitimate browser automation feature, this capability poses a significant risk for data exfiltration if the AI agent is compromised via prompt injection, enabling it to steal sensitive session data. The installation process also involves running `npm install -g` and loading an unpacked Chrome extension, which are common vectors for supply chain attacks, though not inherently malicious in this context. The primary concern is the powerful and potentially abusable `chrome_network_request` tool described in `SKILL.md`.
能力评估
Purpose & Capability
The name and description (Chrome automation via MCP) align with the SKILL.md: it instructs the user to install an MCP native bridge and a Chrome extension, configure an MCP client, and then the skill exposes tools to navigate, click, take screenshots, read history/bookmarks, and monitor network requests. None of the requested items are unrelated to browser automation.
Instruction Scope
The SKILL.md stays within the browser-automation scope but explicitly instructs installing a global npm package and a Chrome extension and describes actions that access highly sensitive browser state (cookies, login sessions, history, bookmarks, network captures). The runtime instructions don't explicitly require explicit per-action user confirmation or limit which pages/actions are allowed, so the agent—if invoked—could perform sensitive operations consistent with the skill but with high privacy impact.
Install Mechanism
This is an instruction-only skill (no install spec in the registry), but the instructions tell users to run `npm install -g mcp-chrome-bridger` and load a Chrome extension from a GitHub Releases zip. The install sources referenced (npm and GitHub Releases) are standard hosts, but installing global npm packages and unpacked browser extensions carries risk if the external project is untrusted. The registry metadata lacks a homepage and the listed owner is not human-readable, so verify the referenced projects before installing.
Credentials
The skill does not request environment variables or credentials, which is proportionate. However, its functionality necessarily reads and uses browser cookies, sessions, history, bookmarks, console output, and network traffic—all highly sensitive data. That sensitivity is legitimate for this type of skill but warrants explicit user awareness and consent before use.
Persistence & Privilege
The skill does not request always: true and does not claim elevated platform privileges. It does require installing local components (npm package + Chrome extension) which will persist until uninstalled; this is normal for a browser-bridging tool but is a form of persistence on the user's machine.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install browser-automation-2
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /browser-automation-2 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v0.1.0
- Initial release of the browser-automation skill for controlling Chrome with MCP protocol. - Allows automation of Chrome browser tasks such as navigation, clicks, form filling, screenshots, and content extraction. - Supports searching browsing history, managing bookmarks, and monitoring network requests in your existing browser session. - Works directly with your current Chrome profile and login sessions; no need for a separate browser instance. - Includes easy installation instructions and compatibility with multiple MCP clients.
元数据
Slug browser-automation-2
版本 0.1.0
许可证
累计安装 3
当前安装数 3
历史版本数 1
常见问题

Browser Automation 是什么?

Control Chrome browser with AI using MCP protocol. Use when users want to automate browser tasks, take screenshots, fill forms, click elements, navigate page... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 709 次。

如何安装 Browser Automation?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install browser-automation-2」即可一键安装,无需额外配置。

Browser Automation 是免费的吗?

是的,Browser Automation 完全免费(开源免费),可自由下载、安装和使用。

Browser Automation 支持哪些平台?

Browser Automation 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Browser Automation?

由 femto(@femto)开发并维护,当前版本 v0.1.0。

💬 留言讨论