mcp-vision MCP Server

What does “mcp-vision” MCP Server do?

The “mcp-vision” MCP Server is a Model Context Protocol (MCP) server that exposes HuggingFace computer vision models—such as zero-shot object detection—as tools to enhance the vision capabilities of large language or vision-language models. By connecting AI assistants with powerful computer vision models, mcp-vision enables tasks like object detection and image analysis directly within development workflows. This allows LLMs and other AI clients to query, process, and analyze images programmatically, making it easier to automate, standardize, and extend vision-based interactions in applications. The server is suited for both GPU and CPU environments and is designed for easy integration with popular AI platforms.

List of Prompts

No specific prompt templates are mentioned in the documentation or repository files.

List of Resources

No explicit MCP resources are documented or listed in the repository.

List of Tools

• locate_objects
  Detect and locate objects in an image using one of the zero-shot object detection pipelines available through HuggingFace. Inputs include the image path, a list of candidate labels, and an optional model name. Returns a list of detected objects in standard format.

• zoom_to_object
  Zoom into a specific object in an image by cropping the image to the bounding box of the object with the best detection score. Inputs include the image path, a label to find, and an optional model name. Returns a cropped image or None.

Use Cases of this MCP Server

• Automated Object Detection in Images
  Developers can use mcp-vision to programmatically detect and locate objects in images, streamlining tasks like image tagging, content moderation, and visual search.
• Vision-based Workflow Automation
  Integrate object detection into larger workflows, such as sorting images by content, automating report generation based on detected items, or enhancing accessibility tools.
• Interactive Image Exploration
  AI assistants can help users zoom in on specific objects within images, aiding tasks like quality inspection, medical imaging analysis, or product identification.
• Augmenting AI Agents with Visual Capabilities
  LLMs can reason about and act on visual data, allowing for richer multimodal interactions and context-aware responses in applications like chatbots, digital assistants, and research tools.

How to set it up

Windsurf

No setup instructions for Windsurf are provided in the repository.

Claude

1. Prerequisites:
   Ensure you have Docker installed and, if using a GPU, an NVIDIA-enabled environment.
2. Build or Use Docker Image:
   • Build locally:

     git clone git@github.com:groundlight/mcp-vision.git
     cd mcp-vision
     make build-docker
     

   • Use public image (optional): No build required.
3. Edit Configuration:
   Open claude_desktop_config.json and add the following under mcpServers:
   • For GPU:

     "mcpServers": {
       "mcp-vision": {
         "command": "docker",
         "args": ["run", "-i", "--rm", "--runtime=nvidia", "--gpus", "all", "mcp-vision"],
         "env": {}
       }
     }
     

   • For CPU:

     "mcpServers": {
       "mcp-vision": {
         "command": "docker",
         "args": ["run", "-i", "--rm", "mcp-vision"],
         "env": {}
       }
     }
     

   • For public image (beta):

     "mcpServers": {
       "mcp-vision": {
         "command": "docker",
         "args": ["run", "-i", "--rm", "--runtime=nvidia", "--gpus", "all", "groundlight/mcp-vision:latest"],
         "env": {}
       }
     }
     

4. Save and Restart:
   Save the configuration and restart Claude Desktop.
5. Verify Setup:
   Ensure that mcp-vision is available as an MCP server in the Claude Desktop UI.

Securing API Keys

• No API key requirements or examples are provided in the documentation.

Cursor

No setup instructions for Cursor are provided in the repository.

Cline

No setup instructions for Cline are provided in the repository.

How to use this MCP inside flows

Using MCP in FlowHunt

To integrate MCP servers into your FlowHunt workflow, start by adding the MCP component to your flow and connecting it to your AI agent:

Click on the MCP component to open the configuration panel. In the system MCP configuration section, insert your MCP server details using this JSON format:

{
  "mcp-vision": {
    "transport": "streamable_http",
    "url": "https://yourmcpserver.example/pathtothemcp/url"
  }
}

Once configured, the AI agent is now able to use this MCP as a tool with access to all its functions and capabilities. Remember to change “mcp-vision” to whatever the actual name of your MCP server is and replace the URL with your own MCP server URL.

Overview

Roots Support: Not mentioned

Overall, mcp-vision provides useful, direct integration with HuggingFace vision models but lacks documentation on resources, prompt templates, or advanced MCP features like roots or sampling. Its setup is well-documented for Claude Desktop but not for other platforms.

Our opinion

mcp-vision is a focused and practical MCP server for adding visual intelligence to AI workflows, especially in environments that support Docker. Its primary strengths are its clear tool offerings and straightforward setup for Claude Desktop, but it would benefit from richer documentation, especially around resources, prompt templates, and support for additional platforms and advanced MCP features.

MCP Score