VisionVisor

An MCP (Model Context Protocol) server that proxies vision requests to an external OpenAI-compatible LLM.

[!IMPORTANT] This project is built entirely on Vibe. High signal, low friction, pure flow.

Features

• Multi-Source Support: Accepts image URLs, local file paths, or Base64 strings.
• OpenAI Compatible: Works with any API that follows the OpenAI Vision API specification (GPT-4o, Claude 3.5 Sonnet, etc.).
• WSL Support: Automatic path mapping for WSL environments.
• MCP Tool: Exposes the ask_vision_model tool to other LLMs.

Quick Start

1. Install dependencies:

   npm install

2. Configure environment: Copy .env.example to .env (already done if you ran the init) and fill in your keys:

   OPENAI_API_KEY=your_key_here
   OPENAI_BASE_URL=https://api.openai.com/v1 # Or your provider's URL
   VISION_MODEL=gpt-4o # Or your preferred vision model

3. Run the server:

   node index.js

Integration with Claude/Desktop

Add this to your MCP settings configuration file:

{
  "mcpServers": {
    "visionvisor": {
      "command": "npx",
      "args": ["-y", "visionvisor"],
      "env": {
        "OPENAI_API_KEY": "your_key_here"
      }
    }
  }
}

Integration with OpenCode

Add this to your opencode.json:

{
  "mcp": {
    "visionvisor": {
      "type": "local",
      "command": ["npx", "-y", "visionvisor"],
      "environment": {
        "OPENAI_API_KEY": "your_key_here"
      },
      "enabled": true
    }
  }
}

Tool Usage: ask_vision_model

Arguments:

• prompt (string): Your question about the image. Pro-tip: Since vision models are stateless, include all necessary context in this prompt.
• image_source (string): Image URL, local file path, or Base64 string.

License

WTFPL - Do What the Fuck You Want to Public License.