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:

1. CSS injection hides common banners (OneTrust, Cookiebot, Intercom, Drift, Zendesk, etc.) before the page loads
2. autoconsent (via DuckDuckGo's database) clicks through 100+ known consent platforms
3. 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