eyeballs
MCP server for visual monitoring: take screenshots of URLs and detect visual changes against stored baselines.
README
eyeballs
Visual monitoring for AI agents and humans. Take screenshots, detect visual changes, track what matters.
Ships as both a CLI tool and an MCP server in one package.
Install
npm install -g eyeballs-cli
This installs Chromium automatically (~400MB on first install).
CLI Usage
# Take a screenshot
eyeballs screenshot https://example.com
eyeballs screenshot https://example.com --viewport 1440x900 -o homepage.png
# Check for visual changes (captures baseline on first run)
eyeballs check https://example.com
eyeballs check https://example.com --threshold 10 --region 0,100,1280,500
# Re-check (diffs against baseline)
eyeballs check https://example.com
# Accept current state as new baseline
eyeballs check https://example.com --reset
# List watched URLs
eyeballs list
# Remove a watch
eyeballs remove <id>
MCP Server
eyeballs works as an MCP server for AI agents (Claude Desktop, Claude Code, Cursor, etc.).
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"eyeballs": {
"command": "eyeballs-mcp"
}
}
}
Claude Code
claude mcp add eyeballs eyeballs-mcp
Tools
| Tool | Description |
|---|---|
screenshot |
Take a screenshot of a URL. Returns the image directly. |
check_url |
Check a URL for visual changes against a stored baseline. |
list_watches |
List all monitored URLs and their status. |
remove_watch |
Remove a watch and its screenshots. |
Example: screenshot
screenshot({ url: "https://example.com", viewport: { width: 1440, height: 900 } })
Returns the screenshot as an image the agent can see, plus metadata (dimensions, load time).
Example: check_url
check_url({ url: "https://example.com", threshold: 5, region: { x: 0, y: 100, width: 1280, height: 500 } })
First call captures a baseline. Subsequent calls diff against it and report the percentage of pixels that changed. Use reset_baseline: true to accept the current state.
Overlay Handling
eyeballs automatically dismisses cookie consent banners, chat widgets, and other overlays before taking screenshots. Three layers:
- CSS injection hides common banners (OneTrust, Cookiebot, Intercom, Drift, Zendesk, etc.) before the page loads
- autoconsent (via DuckDuckGo's database) clicks through 100+ known consent platforms
- Brute-force dismiss finds and clicks remaining "Agree"/"Accept" buttons and removes overlay elements
No configuration needed. Works on most sites out of the box.
How It Works
- Screenshots via Playwright (headless Chromium)
- Diffing via pixelmatch (deterministic pixel comparison, no AI needed)
- Overlay removal via CSS injection + autoconsent + button clicking
- Storage at
~/.eyeballs/(baselines + screenshots as PNG files) - Threshold default 5%, configurable per watch to reduce noise from dynamic content
- Region crop to monitor just the part of the page you care about
Requirements
- Node.js 18+
- ~400MB disk for Chromium (installed automatically)
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。