@ytzry/image-recognition-mcp

MCP server for image recognition using a vision AI model. Supports OCR, scene description, table/chart reading, diagram analysis, and question answering about image content.

Installation

npm install -g @ytzry/image-recognition-mcp

Or run directly with npx (no install needed):

{
  "mcpServers": {
    "image-recognition": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@ytzry/image-recognition-mcp"],
      "env": {
        "VISION_KEY": "your-api-key",
        "VISION_ENDPOINT": "https://your-vision-api-endpoint/v1/chat/completions",
        "VISION_MODEL": "your-model-name"
      }
    }
  }
}

Environment Variables

| Variable | Required | Description | |---|---|---| | VISION_KEY | Yes | API key for the vision model provider | | VISION_ENDPOINT | Yes | OpenAI-compatible chat completions endpoint | | VISION_MODEL | Yes | Vision model name |

Local usage

{
  "mcpServers": {
    "image-recognition": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/image-recognition-mcp/server.mjs"],
      "env": {
        "VISION_KEY": "...",
        "VISION_ENDPOINT": "...",
        "VISION_MODEL": "..."
      }
    }
  }
}

Tool

recognize_image

Analyzes an image using the configured vision AI model.

Parameters:

• image_path (string, optional) — Absolute path to a local image file (jpg, jpeg, png, gif, webp)
• image_url (string, optional) — Publicly accessible image URL
• prompt (string, optional) — Specific instruction. Examples: "OCR all text", "Describe this scene", "Read the table". Defaults to general description + text extraction.

At least one of image_path or image_url is required.

License

MIT