MCP OpenVision

MCP OpenVision is an image analysis MCP server that lets you analyze images using OpenRouter vision models. It exposes a simple, programmable interface so your AI assistants can request detailed insights from images, making it easy to build vision-enabled capabilities into chatbots, assistants, and workflows.

How to use

You run the MCP OpenVision server locally and connect to it with an MCP client. Use the client to send image_analysis requests that describe the image and specify what you want the model to do. The server accepts an image input (URL, file path, or base64 data) and a descriptive query, and returns a structured response with the analysis results.

How to install

Prerequisites: you need Python installed on your machine. You will also use the UV runtime for convenient execution of the server.

# Install the server with Python and UV runtime
# Prerequisites: Python installed on your system

# Install the server package
uv pip install mcp-openvision

# Or install with plain pip if you prefer
pip install mcp-openvision

# Run the server locally for development
export OPENROUTER_API_KEY="your_openrouter_api_key"
python -m mcp_openvision

Additional configuration and usage notes

OpenRouter API key and default model control how the server delegates vision tasks. You must provide an API key to access vision models, and you can set a default model to use when none is specified in requests.

{
  "mcpServers": {
    "openvision": {
      "command": "uvx",
      "args": ["mcp-openvision"],
      "env": {
        "OPENROUTER_API_KEY": "YOUR_OPENROUTER_API_KEY",
        "OPENROUTER_DEFAULT_MODEL": "anthropic/claude-3-sonnet"
      }
    }
  }
}

Testing with MCP Inspector

Quickly verify your setup by testing with the MCP Inspector tool. This allows you to exercise the image_analysis endpoint through the MCP OpenVision server and inspect the results.

npx @modelcontextprotocol/inspector uvx mcp-openvision

Running locally for development

To run the server directly from your development environment, set the required API key and start the module that provides the MCP endpoint.

export OPENROUTER_API_KEY="your_api_key"
python -m mcp_openvision

Available tools

image_analysis

Analyze images using vision models by sending an image input and a query describing the task. Supports image input as a URL, file path, or base64 data, and returns structured results including the model's analysis and any requested formatting.