README.md

metadata

title: medpy MCP Service
emoji: 🔧
colorFrom: blue
colorTo: indigo
sdk: docker
pinned: false
license: mit

medpy MCP Service

This deployment package exposes MedPy medical-image processing capabilities through FastMCP. It supports local stdio usage (Claude Desktop / CLI) and HTTP transport for Docker/HuggingFace Spaces.

Available tools

• health_check — dependency and adapter status
• list_modules — list loaded MedPy modules
• list_symbols — inspect public module symbols
• image_info — inspect shape/dtype/spacing from an image
• anisotropic_diffusion — denoise and save image
• otsu_segmentation — threshold and save binary mask
• compute_overlap_metrics — Dice/Jaccard/Precision/Recall
• bounding_box — non-zero mask bounding box

Local stdio usage

Run from the deployment root:

cd medpy/mcp_output
export MCP_TRANSPORT=stdio
python start_mcp.py

Or:

python medpy/mcp_output/mcp_plugin/main.py

HTTP usage (Docker / HF Spaces)

Docker entrypoint runs:

python medpy/mcp_output/start_mcp.py

with environment:

• MCP_TRANSPORT=http
• MCP_PORT=7860

The MCP HTTP endpoint is available at:

• https://<host>/mcp

Build and run with scripts

• Bash: ./run_docker.sh
• PowerShell: ./run_docker.ps1

Both scripts read port mapping from port.json (fixed to 7860 for HF Spaces compatibility).