loading…
Search for a command to run...
loading…
Identifies high-risk patient cohorts and generates care gap closure plans using a synthetic OMOP-like dataset. It provides clinical decision support tools for h
Identifies high-risk patient cohorts and generates care gap closure plans using a synthetic OMOP-like dataset. It provides clinical decision support tools for homecare management through a DuckDB-backed FastMCP server.
FastMCP server that powers the TopGun HomeCare demo Step 1 agents. It exposes tools to surface the high‑risk cohort and produce the Step 1b care gap closure plan from the synthetic OMOP-like dataset.
/Users/mdnasir/Documents/proj/TopGun/code/homecare-cohort-mcpconda create -n homecare-mcp python=3.11 -y
conda activate homecare-mcp
pip install -r requirements.txt
The synthetic CSVs live in ../synthetic_data. Rebuild the DuckDB file whenever the CSVs change:
cd /Users/mdnasir/Documents/proj/TopGun/code/homecare-cohort-mcp
python -c "from db import ensure_database; ensure_database(force_rebuild=True)"
This creates/overwrites data/homecare.duckdb and materializes helper views (latest_sbp, latest_eye_exam, etc.).
cd /Users/mdnasir/Documents/proj/TopGun/code/homecare-cohort-mcp
uvicorn server:app --reload --port 8010
The Streamable HTTP transport is available at http://127.0.0.1:8010/mcp.
Using the MCP CLI (from the same conda env):
# Identify Step 1a cohort
mcp run server.py:mcp --call get_highrisk_cohort --data '{"limit": 6}'
# Build Step 1b care gap plan
mcp run server.py:mcp \
--call care_gap_closure_plan \
--data '{"patient_ids": ["PAT-00042", "PAT-00058"]}'
Or use MCP Inspector (mcp dev server.py:mcp) to interactively inspect Markdown and structured JSON.
python smoke_test.py
(Ensures both Step 1 tools execute and return non-empty results.)
python -c "from db import ensure_database; ensure_database()"
MCP_ALLOWED_HOSTS / MCP_ALLOWED_ORIGINS and future API keys (CMS, HDI, Medical Research).homecare-cohort-mcp/
├── api/index.py # Vercel entrypoint
├── data/homecare.duckdb # Generated DuckDB file (ignored by default)
├── db.py # DuckDB loader + helper views
├── requirements.txt
├── server.py # FastMCP server with Step 1 tools
├── smoke_test.py # Regression script
└── vercel.json
Добавь это в claude_desktop_config.json и перезапусти Claude Desktop.
{
"mcpServers": {
"homecare-cohort-mcp": {
"command": "npx",
"args": []
}
}
}