Image Understanding MCP

An MCP server that uses the Zhipu AI GLM vision models to analyze and describe images.

Features

• Supports multiple GLM vision models (glm-4v-plus, glm-4v, glm-4v-flash)
• Flexible generation controls (temperature, max tokens, etc.)
• Robust error handling with clear messaging
• TypeScript-based implementation with strong typing

Environment Variables

• ZHIPU_API_KEY or GLM_API_KEY: Required, Zhipu AI API key
• GLM_VISION_MODEL: Optional, override the default model (defaults to glm-4v-plus)

Install & Build

npm install
npm run build

Run

# Run the compiled entry
npm start
# Or invoke the CLI wrapper
image-understanding-mcp

MCP Client Configuration

Example entry for Claude Desktop:

{
  "mcpServers": {
    "image-understanding": {
      "command": "node",
      "args": ["/absolute/path/to/this/project/dist/index.js"],
      "env": {
        "ZHIPU_API_KEY": "your-zhipu-key"
      }
    }
  }
}

Exposed Tool

image_understanding

Analyze an image via the GLM vision model suite.

Parameters:

• imageUrl (required): Image URL
• prompt (optional): Instruction for the analysis, defaults to “Describe this image in detail.”
• model (optional): Model id, default glm-4v-plus
• temperature (optional): 0-1, default 0.7
• maxTokens (optional): 1-4096, default 1024

Supported Models:

• glm-4v-plus: Highest fidelity, best for complex scenes (max 4096 tokens)
• glm-4v: Balanced performance/cost (max 2048 tokens)
• glm-4v-flash: Fastest inference for simple analysis (max 1024 tokens)

Project Layout

src/
├── index.ts          # MCP server entry
├── config.ts         # Model options
├── types.ts          # Shared types
└── tool.ts           # Tool implementation

Compatibility

• Node.js >= 18
• @modelcontextprotocol/sdk v1