Server Browser Automation

在无桌面 Ubuntu 服务器上通过 XFCE 虚拟桌面和 Chrome 远程调试实现浏览器自动化及持久登录状态管理。

Security Scan

The skill's purpose (setting up a VNC/XFCE + Chrome remote-debugging environment for OpenClaw) matches what the files do, but the installer has user/root handling and path assumptions that don't align with the documentation and could lead to misconfiguration or unintended root-level setup — review and test before installing on a production host.

Purpose & Capability

The skill's name/description (server-side browser automation with persistent Chrome user data) aligns with the SKILL.md and install.sh actions (install XFCE, VNC, Chrome, configure OpenClaw). However the installer forces root execution and then configures files in the invoking account's home using whoami; in practice this script will configure root's home (or otherwise mis-target user home), which contradicts many examples in SKILL.md that reference a non-root user's /home/your_user paths. This mismatch is a coherence issue that can cause unintended root-scoped configuration.

Instruction Scope

SKILL.md stays within the stated purpose: it instructs installing desktop/VNC, starting Chrome with remote debugging, and doing a manual login so cookies persist. It does instruct the operator to log in to third-party sites (Douyin, Weibo, Twitter) and rely on persistent user-data directories; this gives the agent/local processes access to session cookies and site data, which is intended but should be understood by the user. No instructions reference exfiltration endpoints or unrelated system data.

Install Mechanism

There is no packaged install spec; the provided install.sh performs apt installs and adds Google's Chrome apt repository via dl.google.com (official). Using apt-key add is deprecated but not unusual. The script needs root/sudo and installs system packages — appropriate for this task but high-impact because it modifies system packages and services.

Credentials

The skill declares no environment variables or external credentials, which is consistent with its function. The concern is the required privilege level: the installer demands root/sudo and then uses whoami to derive the target user/home, causing the configuration to target root rather than an unprivileged user in common usage. This is disproportionate in the sense of unexpected scope for a user who intended to install for a regular account. No other unrelated credentials are requested.

Persistence & Privilege

The skill does not request always:true and does not modify other skills. It does, however, perform system-wide changes (apt installs, adding repo, creating /root or a user's ~/.vnc and ~/.openclaw config) which are persistent and require root. That level of system persistence is expected for installing desktop/VNC on a server but should be accepted consciously by the operator.

What to consider before installing

This package appears to implement what it claims (VNC/XFCE + Chrome remote debugging for OpenClaw), but there are important practical issues to review before installing: - The install.sh requires root/sudo and then uses whoami to decide which home directory to configure. If you run it with sudo it will configure root's home (~root) rather than your normal user. Prefer to run as root only if you intend the service for root, or modify the script to use SUDO_USER (or pass an explicit TARGET_USER) so configuration goes into the intended account. - The start-chrome.sh template uses a literal $(whoami) within a single-quoted here-doc; verify the produced script's paths match the actual user who will run it. - The installer will add system packages and the Google Chrome apt repo and start services that listen on network ports (VNC 5901, Chrome remote debugging 18800). Do not expose VNC or Chrome debugging ports to the public internet — use SSH tunnels, noVNC behind authentication, or firewall rules. - Back up ~/.openclaw/openclaw.json if present (the script does create a backup) and inspect changes the script will make. Consider running this in a disposable VM/container first rather than on a production server. - If you need the skill to run for a non-root user, edit install.sh to set VNC_USER from SUDO_USER (fallback to a provided username) and ensure file ownerships and paths are correct. If you want, I can point out the exact lines in install.sh to change (SUDO_USER handling and here-doc quoting) and draft a safer variant of the installer that targets a specific non-root user and doesn't rely on apt-key add.

Server Browser Automation Skill

在无桌面服务器上实现 OpenClaw 浏览器自动化的完整解决方案。

📖 描述

Server Browser Automation 是一个帮助你在无桌面 Ubuntu 服务器上配置和使用 OpenClaw 浏览器自动化功能的 Skill。

解决的核心问题：

• ❌ 无桌面环境无法显示浏览器
• ❌ Chrome 远程调试配置复杂
• ❌ AI 无法操作需要登录的网站
• ❌ Cookie 无法持久化保存

提供的解决方案：

• ✅ XFCE + VNC 虚拟桌面
• ✅ Chrome 远程调试配置
• ✅ 个人资料模式（Cookie 持久化）
• ✅ AI 自然语言操作浏览器

🎯 适用场景

场景	说明
数据采集	爬取抖音、微博、Twitter 等需要登录的网站
自动化测试	自动测试网页功能、表单提交
定时任务	定期抓取数据、监控价格变化
批量操作	多账号管理、批量发布内容

📦 安装

前置要求

• Ubuntu Server 20.04+（无桌面版本）
• OpenClaw 2026.3.13+
• Node.js 22+
• Docker（可选，用于 noVNC）
• root 或 sudo 权限

一键安装脚本

# 1. 克隆 Skill
git clone https://github.com/YOUR_USERNAME/server-browser-automation.git
cd server-browser-automation

# 2. 运行安装脚本
./install.sh

# 3. 验证安装
openclaw browser --browser-profile openclaw status

手动安装

# 1. 安装 XFCE 桌面
sudo apt-get update
sudo apt-get install -y xfce4 xfce4-goodies

# 2. 安装 VNC 服务器
sudo apt-get install -y tigervnc-standalone-server tigervnc-common

# 3. 设置 VNC 密码
vncpasswd

# 4. 启动 VNC
vncserver :1 -geometry 1920x1080 -depth 24

# 5. 安装 Chrome
wget -q -O - https://dl.google.com/linux/linux_signing_key.pub | sudo apt-key add -
sudo sh -c 'echo "deb [arch=amd64] http://dl.google.com/linux/chrome/deb/ stable main" >> /etc/apt/sources.list.d/google-chrome.list'
sudo apt-get update
sudo apt-get install -y google-chrome-stable

# 6. 配置 OpenClaw
# 编辑 ~/.openclaw/openclaw.json，添加 browser 配置（见下方配置章节）

# 7. 启动 Chrome
export DISPLAY=:1
google-chrome \
  --remote-debugging-port=18800 \
  --no-sandbox \
  --disable-gpu \
  --disable-dev-shm-usage \
  --user-data-dir=~/.config/openclaw-browser-openclaw \
  > /tmp/chromium.log 2>&1 &

# 8. 重启 Gateway
openclaw gateway restart

⚙️ 配置

OpenClaw 配置

编辑 ~/.openclaw/openclaw.json：

{
  "browser": {
    "enabled": true,
    "defaultProfile": "openclaw",
    "headless": false,
    "remoteDebuggingPort": 18800,
    "userDataDir": "/home/your_user/.config/openclaw-browser-openclaw",
    "profiles": {
      "openclaw": {
        "cdpPort": 18800,
        "color": "#FF4500",
        "userDataDir": "/home/your_user/.config/openclaw-browser-openclaw"
      }
    }
  }
}

VNC 配置

创建 ~/.vnc/xstartup：

#!/bin/bash
unset SESSION_MANAGER
unset DBUS_SESSION_BUS_ADDRESS
exec startxfce4

chmod +x ~/.vnc/xstartup

🚀 使用

基本命令

# 检查浏览器状态
openclaw browser --browser-profile openclaw status

# 查看标签页
openclaw browser --browser-profile openclaw tabs

# 打开网页
openclaw browser --browser-profile openclaw open https://www.example.com

# 截图
openclaw browser --browser-profile openclaw screenshot page.png

# 执行 JavaScript
openclaw browser --browser-profile openclaw eval "document.title"

自然语言指令

直接用自然语言告诉 AI：

帮我打开抖音
用浏览器访问 https://www.douyin.com
看看百度的标题是什么
帮我登录抖音，然后爬取前 20 个视频
截图当前页面
提取这个页面的所有视频链接

首次登录流程

1. 启动 Chrome（见安装步骤）
2. 用 VNC 查看器连接：服务器 IP:5901
3. 在 Chrome 中访问目标网站（如抖音）
4. 完成登录（扫码 + 验证码）
5. 保持 Chrome 运行
6. 后续 AI 自动操作（Cookie 已保存）

自动化示例

抖音数据爬取

# 打开抖音（首次需要人工登录）
openclaw browser --browser-profile openclaw open https://www.douyin.com

# 登录后，后续可以直接爬取
# AI 会自动提取数据并保存

价格监控

# 每天检查价格
openclaw browser --browser-profile openclaw open https://item.jd.com/xxx.html

# AI 提取价格并保存

自动化测试

# 测试登录功能
openclaw browser --browser-profile openclaw open https://example.com/login

# AI 自动填写表单并提交

🔧 故障排查

Chrome 未启动

# 检查进程
ps aux | grep chrome

# 查看日志
cat /tmp/chromium.log

# 重新启动
export DISPLAY=:1
google-chrome --remote-debugging-port=18800 --no-sandbox --user-data-dir=~/.config/openclaw-browser-openclaw &

OpenClaw 无法连接

# 检查 CDP 端口
curl http://127.0.0.1:18800/json/list

# 检查配置
cat ~/.openclaw/openclaw.json | jq '.browser'

# 重启 Gateway
openclaw gateway restart

VNC 无法连接

# 检查 VNC 进程
ps aux | grep vnc

# 重启 VNC
vncserver -kill :1
vncserver :1 -geometry 1920x1080 -depth 24

Cookie 失效

# 删除旧 Cookie
rm -rf ~/.config/openclaw-browser-openclaw

# 重新启动 Chrome，重新登录
export DISPLAY=:1
google-chrome --remote-debugging-port=18800 --no-sandbox --user-data-dir=~/.config/openclaw-browser-openclaw &

📊 架构说明

┌─────────────────────────────────────────────────────────┐
│                    用户（本地电脑）                       │
│                  VNC 查看器 / noVNC                       │
└────────────────────┬────────────────────────────────────┘
                     │ VNC (端口 5901)
                     ▼
┌─────────────────────────────────────────────────────────┐
│              服务器（Ubuntu 无桌面）                       │
│  ┌─────────────────────────────────────────────────┐    │
│  │              XFCE 虚拟桌面 (:1)                   │    │
│  │  ┌─────────────────────────────────────────┐    │    │
│  │  │          Google Chrome                   │    │    │
│  │  │  远程调试端口：18800                      │    │    │
│  │  │  User Data: Cookie 持久化                 │    │    │
│  │  └─────────────────────────────────────────┘    │    │
│  └─────────────────────────────────────────────────┘    │
│                     │ CDP                                 │
│                     ▼                                     │
│  ┌─────────────────────────────────────────────────┐    │
│  │           OpenClaw Gateway                       │    │
│  │           端口：18789                            │    │
│  └─────────────────────────────────────────────────┘    │
└─────────────────────────────────────────────────────────┘
                     │ WebSocket
                     ▼
┌─────────────────────────────────────────────────────────┐
│                    AI 助手                               │
│              （飞书 / Telegram / Discord）                │
└─────────────────────────────────────────────────────────┘

🎓 进阶技巧

1. 保存登录状态

# Cookie 自动保存在 userDataDir
# 无需手动保存，下次启动 Chrome 自动加载

2. 多账号管理

# 为不同账号创建不同的 userDataDir
google-chrome --remote-debugging-port=18801 --user-data-dir=~/.config/chrome-account1 &
google-chrome --remote-debugging-port=18802 --user-data-dir=~/.config/chrome-account2 &

3. 定时任务

# 添加 cron 任务
crontab -e

# 每天 9 点执行
0 9 * * * openclaw browser --browser-profile openclaw open https://www.douyin.com

4. 数据导出

# AI 自动提取数据并保存为 JSON/CSV
# 示例：提取抖音视频数据
openclaw browser --browser-profile openclaw eval "
  Array.from(document.querySelectorAll('.video-card')).map(card => ({
    title: card.querySelector('.video-title')?.innerText,
    author: card.querySelector('.author-name')?.innerText,
    plays: card.querySelector('.play-count')?.innerText
  }))
"

⚠️ 注意事项

安全

• 不要在生产服务器上启用远程调试（仅限开发/测试环境）
• VNC 密码要设置复杂
• 防火墙只允许信任的 IP 访问 VNC 端口

性能

• Chrome 会占用较多内存（建议服务器至少 2GB 内存）
• 定期清理 userDataDir（避免 Cookie 缓存过大）
• 使用 --disable-gpu 减少 GPU 依赖

反爬

• 控制请求频率（避免被封 IP）
• 使用代理 IP（如需大规模爬取）
• 遵守网站 robots.txt

📚 相关资源

• OpenClaw 官方文档
• Chrome DevTools Protocol
• VNC 官方文档
• XFCE 官方文档

🤝 贡献

欢迎提交 Issue 和 Pull Request！

📄 许可证

MIT License

👥 作者

• 开发者
• OpenClaw 社区

🙏 致谢

感谢 OpenClaw 团队提供的优秀框架！

---

Happy Automating! 🚀