This MCP server implements the Model Context Protocol for image processing tasks using Ultralytics models. It exposes tools to detect objects, segment objects, segment entire images, and estimate human poses from images. The server supports both stdio and SSE transport modes, allowing you to run it locally or connect via HTTP SSE endpoints. Typical usage is to start the server with the python server.py command and then interact with the provided tools through the MCP client, feeding image URLs or local file paths and receiving structured JSON results for each operation.

Available tools include:

• detect_objects: runs YOLOv10-based object detection on an input image and returns an array of detections with bounding boxes, confidence scores, and class labels.
• segment_objects: uses YOLOv8-based segmentation to identify and segment individual objects in an image, returning bounding boxes, class labels, and confidence scores.
• segment_image: applies Ultralytics SAM to segment the entire image, returning regions with bounding boxes, area, and confidence scores.
• estimate_pose: detects human poses in an image with YOLOv8 and returns keypoints and confidence scores for detected poses.

To interact, you can start the server and connect via the MCP client, sending requests that specify the tool you want to use along with the image_url (or file path). The server expects valid image paths or URLs and requires the corresponding model weights to be present in the checkpoints directory (yolov10b.pt, yolov8n-seg.pt, yolov8n-pose.pt, sam_b.pt).

Prerequisites:

• Python 3.8+ and pip
• Git
• Network access to download dependencies and model weights

Step-by-step:

1. Clone the repository git clone https://github.com/MRonaldo-gif/mronaldo-gif-mcp-server-cvdlt.git cd mronaldo-gif-mcp-server-cvdlt

2. Create and activate a virtual environment (optional but recommended) python -m venv venv

   Windows

   venv\Scripts\activate

   macOS/Linux

   source venv/bin/activate

3. Install dependencies pip install -r requirements.txt

4. Prepare model weights

   • Download required weights and place them in the checkpoints directory: yolov10b.pt, yolov8n-seg.pt, yolov8n-pose.pt, sam_b.pt
   • Weights can be obtained from the official model docs and should be placed under ./checkpoints

5. Start the server python server.py

   Or for SSE mode (if supported by the script):

   python server.py sse 8080

6. Verify the server is running by connecting with an MCP client or by checking logs for startup messages.

Tips and tips:

• Ensure all required weights are present in ./checkpoints before starting the server to avoid runtime errors.
• The server supports both local image paths (file:// or relative paths) and network URLs (http/https). Use image URLs for remote images.
• If behind a slow network or with large models, consider increasing system resources or using a local lightweight environment to improve startup time.
• The provided QuickStart includes a Claude Desktop integration example; you can adapt the mcpServers.json structure to point to the SSE endpoint for Claude or other clients.
• For troubleshooting, check that the Python environment has the necessary dependencies and that the required ports are not blocked by a firewall.