Smart visual analysis that handles both websites and images.
Usage
/vision <url> [prompt]
<url>— Website URL OR direct image URL[prompt]— Optional: what to look for (default: comprehensive analysis)
Behavior
| URL Type | What Happens |
|---|---|
Website (https://example.com) |
Playwright navigates → screenshots → analyzes |
| Image (`https://.../*.png | jpg |
Examples
/vision https://example.com
→ Navigates to site, takes screenshot, analyzes layout and design
/vision https://example.com "What colors are used?"
→ Screenshots site, analyzes color palette
/vision https://mockup.png "replicate"
→ Direct image analysis for UI replication
/vision https://diagram.png "Describe the architecture"
→ Direct image analysis
For UI Replication
/vision https://mysite.com "replicate"
Uses the replication prompt: "Describe in detail the layout structure, color style, main components, and interactive elements to facilitate code generation."
Execution Steps
When invoked:
-
Detect URL type:
- Ends in
.png,.jpg,.jpeg(case insensitive) → image - Otherwise → website
- Ends in
-
For images:
- Call
mcp__4_5v_mcp__analyze_imagedirectly - If prompt is "replicate", use specialized replication prompt
- Otherwise use provided prompt or default comprehensive analysis
- Call
-
For websites:
- Call
mcp__playwright__browser_navigatewith the URL - Call
mcp__playwright__browser_screenshotto capture - Call
mcp__4_5v_mcp__analyze_imageon the screenshot - If prompt is "replicate", use specialized replication prompt
- Call
Tools Used
mcp__playwright__browser_navigate— Navigate to websitesmcp__playwright__browser_screenshot— Capture screenshotsmcp__4_5v_mcp__analyze_image— AI vision analysis
Requirements
- Playwright MCP server must be configured
- The
mcp__4_5v_mcp__analyze_imageMCP must be available
Local Images
For local image files, use Read tool directly:
Read: screenshots/mockup.png