Computer Vision MCP Server

# Usage examples (URL only): # just double_flash https://example.com/img.jpg # just double_pro https://example.com/img.jpg # just double_qwen_local https://example.com/img.jpg Qwen/Qwen2.5-VL-7B-Instruct set shell := ["bash", "-cu"] # Double (vision alt+caption + text metadata) double_flash url: python cli/image_metadata.py --image-url "{{url}}" --mode double --config-path configs/double_flash.json double_pro url: python cli/image_metadata.py --image-url "{{url}}" --mode double --config-path configs/double_pro.json # Recommended mix: Flash for vision, Pro for text metadata double_mixed_pro_text url: python cli/image_metadata.py --image-url "{{url}}" --mode double --config-path configs/double_mixed_pro_text.json # Triple (vision alt+caption + vision metadata) triple_flash url: python cli/image_metadata.py --image-url "{{url}}" --mode triple --config-path configs/triple_flash.json triple_pro url: python cli/image_metadata.py --image-url "{{url}}" --mode triple --config-path configs/triple_pro.json # Local Qwen 2.5 VL scenarios # Double: local vision, remote text metadata (Pro) double_qwen_local url qwen_id: python cli/image_metadata.py --image-url "{{url}}" --mode double --caption-backend local --local-vlm-id "{{qwen_id}}" --config-path configs/double_pro.json # Triple: fully local (no OpenRouter) triple_qwen_local url qwen_id: python cli/image_metadata.py --image-url "{{url}}" --mode triple --caption-backend local --metadata-vision-backend local --local-vlm-id "{{qwen_id}}" # Convenience: fixed Qwen variants double_qwen2b_local url: python cli/image_metadata.py --image-url "{{url}}" --mode double --caption-backend local --local-vlm-id "Qwen/Qwen2-VL-2B-Instruct" --config-path configs/double_pro.json triple_qwen2b_local url: python cli/image_metadata.py --image-url "{{url}}" --mode triple --caption-backend local --metadata-vision-backend local --local-vlm-id "Qwen/Qwen2-VL-2B-Instruct" double_qwen7b_local url: python cli/image_metadata.py --image-url "{{url}}" --mode double --caption-backend local --local-vlm-id "Qwen/Qwen2.5-VL-7B-Instruct" --config-path configs/double_pro.json triple_qwen7b_local url: python cli/image_metadata.py --image-url "{{url}}" --mode triple --caption-backend local --metadata-vision-backend local --local-vlm-id "Qwen/Qwen2.5-VL-7B-Instruct" # Install optional local dependencies (transformers, torch, etc.) setup_local: python -m pip install -U pip wheel python -m pip install '.[local]' # Ollama (no transformers) — fully local vision (triple) triple_qwen_ollama url: python cli/image_metadata.py --image-url "{{url}}" --mode triple --caption-backend ollama --metadata-vision-backend ollama --config-path configs/triple_ollama_qwen.json