desktop-vision

Desktop Vision Skill

You have access to DesktopLens MCP tools that let you see and analyze desktop applications outside the browser. Use these tools to capture screenshots, stream windows in real-time, compare UI changes, and manage plugins.

Available Tools

Core Tools

desktoplens_list_windows

List all visible desktop windows. Use this first to find the target window.

• filter (optional): Fuzzy search by window title or app name

desktoplens_screenshot

Capture a screenshot of a specific window.

• window_id or window_title: Target window identifier
• format (optional): png | jpeg | webp (default: png)
• max_width / max_height (optional): Resize constraints
• annotate (optional): Add grid overlay for region reference

desktoplens_watch

Start real-time streaming of a window to a Chrome viewer.

• window_id or window_title: Target window identifier
• fps (optional): Frame rate 0.5–5 (default: 2)
• quality (optional): low | medium | high (default: medium)
• open_browser (optional): Auto-open Chrome viewer (default: true)

desktoplens_stop

Stop streaming session(s).

• session_id (optional): Specific session to stop. Omit to stop all.

desktoplens_status

Get server status, platform info, and active streaming sessions.

desktoplens_compare

Compare two screenshots to detect UI changes.

• before_snapshot_id or before_window_id/before_window_title: Before state
• after_window_id or after_window_title: After state
• highlight_diff (optional): Generate visual diff image
• threshold (optional): Pixel difference sensitivity

Plugin Tools

desktoplens_plugin_search

Search GitHub for DesktopLens plugins.

• query: Search keywords

desktoplens_plugin_install

Install a plugin from a local directory path.

• source: Path to plugin directory

desktoplens_plugin_list

List all installed plugins with their status and tools.

desktoplens_plugin_remove

Remove an installed plugin.

• plugin_name: Name of the plugin to remove

Workflow Patterns

Quick Screenshot

1. Call desktoplens_list_windows to find the target window
2. Call desktoplens_screenshot with the window ID or title
3. Analyze the returned screenshot image

Before/After Comparison

1. Call desktoplens_screenshot to capture the "before" state (note the snapshot_id)
2. Ask the user to make their changes
3. Call desktoplens_compare with the before snapshot ID and current window

Real-time Monitoring

1. Call desktoplens_watch to start streaming
2. The Chrome viewer opens automatically showing the live feed
3. Call desktoplens_status to check active sessions
4. Call desktoplens_stop when monitoring is complete

Best Practices

• Always call desktoplens_list_windows first to verify the target window exists and get its exact ID
• Use window_title for convenience (supports fuzzy matching) or window_id for precision
• For UI analysis, use annotate: true to add a grid overlay that helps reference specific regions
• Set appropriate quality levels: low for quick checks, high for detailed analysis
• Stop streaming sessions when no longer needed to free resources
• Screenshots are automatically compressed for Claude Vision token efficiency