Overview
What is farshid-mcp-imageProcessing?
farshid-mcp-imageProcessing is a comprehensive OpenCV image-processing MCP server for VS Code Copilot Agent Mode (or any MCP client). It exposes ~40 tools across webcam capture, image I/O, transforms, color, filtering, edges, thresholding, morphology, contours/shapes, feature matching, object detection, drawing, composition, template matching, and video processing. It is for developers who want to perform computer vision tasks directly from an MCP client.
How to use farshid-mcp-imageProcessing?
Install with pip install farshid-mcp-imageProcessing, then run farshid-mcp-imageprocessing to start the stdio server. Register it in VS Code by adding a imageProcessing entry to your mcp.json with the command farshid-mcp-imageprocessing and type stdio. Optionally, clone the repo and run from source. For quick testing, use the /cv prompt in Copilot Chat (e.g., /cv take image from webcam and save it as gray scale 240 * 240).
Key features of farshid-mcp-imageProcessing
- Webcam capture, preview, record, and save
- Geometric transforms: resize, crop, rotate, flip, pad
- Color conversion and histogram equalization
- Filtering (Gaussian, median, bilateral) and sharpening
- Edge detection (Canny, Sobel, Laplacian)
- Object detection via Haar cascades (faces, eyes, bodies, QR)
Use cases of farshid-mcp-imageProcessing
- Capture a photo from a webcam and apply grayscale and resize transformations
- Detect faces in an image and draw bounding boxes around them
- Compare two images for differences using absolute difference
- Extract frames from a video file at a specified interval
- Match a template pattern within a larger image
FAQ from farshid-mcp-imageProcessing
What Python version is required?
Python 3.14 or later is required.
Does this server work on a headless server?
Most tools work on headless servers, but webcam_preview opens a real desktop window and will fail without a display.
How do I handle camera permissions on macOS?
The first webcam call triggers a system Camera permission prompt; grant it to the terminal or VS Code process.
What should I avoid when writing custom tools or using this server?
Never use print() in tool functions because stdout is the MCP protocol channel; use sys.stderr (the _log helper in server.py) instead.
Is opencv-contrib-python included?
Yes, the package uses opencv-contrib-python to provide bundled Haar cascades and extra algorithms.
