Vision

Vision MCP Server provides a dedicated toolset to add vision capabilities to models that otherwise lack vision support. The core tool is analyze_image, which can take an image URL, a local file path, or base64-encoded image data, and return a structured analysis from a selected vision model hosted on OpenRouter. You can also query list_models to discover available vision models and their capabilities. Configure the server in your MCP client to route analyze_image and list_models requests through vision-mcp, enabling your assistant to analyze images using models like Claude-3.5 Sonnet or GPT-4 Vision without embedding vision directly into your base model.

Prerequisites:

• Node.js 18.0.0 or higher
• npm 8.0.0 or higher

Install from npm (recommended):

npm install -g @thenomadinorbit/vision-mcp-server

This makes the vision-mcp command globally available as vision-mcp.

Alternatively, install from source (development):

git clone https://github.com/TheNomadInOrbit/vision-mcp-server.git
cd vision-mcp-server
npm install
npm run build
npm install -g .

After installation, ensure vision-mcp is accessible in your shell and configure your MCP client with the provided server settings (see the Configuration section in the README).

Tips and common issues:

• The server requires an OpenRouter API key. Do not run analyze_image without configuring OPENROUTER_API_KEY in the environment or MCP client config.
• The OPENROUTER_MODEL environment variable selects the vision model. Popular options include claude-3-5-sonnet, gpt-4o, and gemini-2.0.
• MAX_IMAGE_SIZE controls the maximum allowed image payload; adjust if you work with very large images.
• If you encounter networking or authentication errors, verify that the API key is valid and that network access to OpenRouter endpoints is not blocked by a firewall.
• Auto-approval settings help streamline workflows by pre-authorizing specific tools; be mindful of the security implications if you enable analyze_image in autoApprove.
• The server uses TypeScript-based code; ensure you are running on Node.js 18+ as noted in the requirements.