What is Image Vision Mcp

image-vision-mcp is an MCP Server that provides descriptions of images by leveraging the capabilities of the Llava language model.

Use cases

Use cases include generating alt text for images, assisting visually impaired users, automating content tagging, and enhancing user interaction in applications that utilize visual data.

How to use

To use image-vision-mcp, install the required Llava model via Ollama, configure the MCP in the claude_desktop_config.json file, and run the server using the command: node src/image-vision-mcp.js --permitted /path/to/dir1 /path/to/dir2.

Key features

Key features include the ability to specify permitted directories for security, support for different models via the --model argument, and integration with the MCP inspector for development.

Where to use

image-vision-mcp can be used in fields such as image processing, content creation, accessibility tools, and any application requiring image description capabilities.

Image Vision MCP

MCP Server that provides descriptions of images.

Requirements

Ollama

You must have Ollama set up, and exposing its LLMs via its API / webserver.

By default, the MCP uses the llava:34b LLM, which you need to have installed and on the same computer that Ollama is running on.

You can install the model via:

ollama run llava:34b

You can also specify a different model installed via Ollama via the –model argument in Claude config.

Installation

In order to install the MCP into Claude, add the following to the claude_desktop_config.json file.

    "image-vision": {
      "command": "npx",
      "args": [
        "-y",
        "node",
        "C:/Users/FOO/src/image-vision-mcp/src/image-vision-mcp.js",
        "--permitted",
        "C:/Users/FOO/mcp",
        "--host",
        "http://192.168.1.238:11434"
      ]
    }

Arguments

• –permitted (required) The paths that the MCP is allowed to access. You can include multiple entries.
• –host (optional) The host and post for the Ollama server. Defaults to http://127.0.0.1:11434
• –model (optional) The model installed into Ollama that should be used for the image vision. Defaults to “llava:34b”.

Usage

To run the server:

node src/image-vision-mcp.js --permitted /path/to/dir1 /path/to/dir2

The --permitted flag is used to specify which directories roots the MCP is allowed to access for security reasons.

Development

You can run in development mode using the MCP inspector:

npx @modelcontextprotocol/inspector node src/image-vision-mcp.js --permitted /Users/FOO/Desktop/mcp/

Questions, Feature Requests, Feedback

If you have any questions, feature requests, need help, or just want to chat, join the discord.

You can also log bugs and feature requests on the issues page.

License

Project released under a MIT License.