Overview
What is MCP Vision Adapter?
MCP Vision Adapter is a universal, open-source MCP (Model Context Protocol) adapter for all CLI projects. It modernizes and connects classic automations, scripts, and detection tools to LLM/agent ecosystems.
How to use MCP Vision Adapter?
Run the YOLOv8 service and the adapter using Docker Compose, or set up a virtual environment, install dependencies, and start both services with Uvicorn. Invoke via HTTP/STDIO or VS Code Copilot Agent mode using endpoints like /execute with tool detect_objects and input parameters.
Key features of MCP Vision Adapter
- Ultralytics YOLOv8 integration for object detection, segmentation, and pose estimation.
- REST & STDIO API for flexible integration.
- Web UI for model management via Streamlit.
- VS Code Copilot Agent mode support.
- Easy Docker Compose setup.
- Video and image analysis capabilities.
Use cases of MCP Vision Adapter
- Turn legacy CLI automations into LLM/agent tools without rewriting code.
- Perform object detection on images and videos directly from agent workflows.
- Hot‑swap detection models on a running service.
- Connect any CLI‑based tool as an agent tool via the MCP protocol.
FAQ from MCP Vision Adapter
What is the MCP protocol version used?
MCP Protocol Version 2024-03-26.
How can I run the project without Docker?
Create a Python virtual environment, install dependencies from yolov8_service/requirements.txt, then start the YOLOv8 service on port 8080 and the adapter on port 3000.
Does MCP Vision Adapter support VS Code Copilot via STDIO?
VS Code Copilot tool execution via STDIO is currently not supported. Use the HTTP API for full integration; this will be fixed in an upcoming update.
What if port 3000 is already in use?
Stop the previous process (CTRL‑C) or run the adapter with --port 3001 and update mcp.json accordingly.
Where are the custom models stored and how to mount them?
Default weights file is /weights/yolov8n.pt. Mount your model folder with docker run -v ${PWD}/models:/weights.
