zai-vision

name	zai-vision
description	Analyze images and videos using Z.AI GLM-4V vision model. Use when the user needs to analyze screenshots or UI images, extract text from images (OCR), diagnose error screenshots, understand technical diagrams or architecture charts, analyze data visualizations, compare two UI screenshots, or generate code from a UI image. Run scripts/analyze.py via Bash.
compatibility	Requires Python 3 and Z_AI_API_KEY environment variable.
metadata	{"author":"z-ai","version":"1.0","api-docs":"https://docs.bigmodel.cn/cn/coding-plan/mcp/vision-mcp-server"}

Z.AI Vision Skill

Analyze images and videos by running scripts/analyze.py with Bash. No MCP configuration needed — set Z_AI_API_KEY and call the script directly.

Prerequisites

export Z_AI_API_KEY=your_api_key

Usage

python3 scripts/analyze.py <image_path_or_url> [--mode <mode>] [--question "custom question"]

Modes

Mode	--mode flag	When to use
General analysis	analyze (default)	Any image, no specific task
OCR / text extraction	ocr	Extract text content from a screenshot
UI to code	ui-to-code	Convert UI screenshot to HTML/CSS code
Error diagnosis	diagnose	Parse error dialogs and suggest fixes
Technical diagram	diagram	Explain architecture/UML/ER diagrams
Chart analysis	chart	Extract insights from graphs and dashboards
UI diff	diff	Compare two screenshots for differences

Examples

# General image analysis
python3 scripts/analyze.py screenshot.png

# Extract text from an image (OCR)
python3 scripts/analyze.py screenshot.png --mode ocr

# Convert UI mockup to code
python3 scripts/analyze.py design.png --mode ui-to-code

# Diagnose an error screenshot
python3 scripts/analyze.py error.png --mode diagnose

# Explain an architecture diagram
python3 scripts/analyze.py arch.png --mode diagram

# Analyze a chart/dashboard
python3 scripts/analyze.py chart.png --mode chart

# Compare two UI screenshots
python3 scripts/analyze.py before.png after.png --mode diff

# Ask a custom question
python3 scripts/analyze.py image.png --question "What color is the button?"

# Use a remote image URL
python3 scripts/analyze.py https://example.com/image.png --mode ocr

Instructions for the Agent

1. Identify which mode fits the user's request using the table above.
2. Locate the image file path (or URL) from the user's message or working directory.
3. Run the script with Bash using the appropriate --mode.
4. Present the output to the user.

For UI diff, pass two image paths: python3 scripts/analyze.py img1.png img2.png --mode diff