Rx Shortage Alternative Finder (MCP)

An MCP server that lets an LLM answer a pharmacist's question: "Drug X is on backorder — what could be considered instead, and are any of those also short?"

When a drug goes into shortage, the obvious therapeutic alternative is often also short — a documented "cascading shortage" (e.g. cisplatin → carboplatin, where the alternative became more scarce than the original). The data to check this is public, but no single tool stitches it together. This server does, in one LLM reasoning pass.

⚠️ Decision-support, not a substitution authority. "Same pharmacologic class" ≠ "clinically interchangeable." This tool surfaces candidates for a licensed professional to evaluate. It does not check route, indication, contraindications, or therapeutic equivalence. See Safety.

✅ Status: v1 — all four tools live, orchestrated, and safety-gated; verified live in Claude Desktop. 15/15 on the chain-success eval.

---

Who it's for & what makes it different

Built for informatics pharmacists and clinical-data teams who operationalize shortage response — people who care about how the data is stitched together, not just the answer. Every suggestion is grounded in real RxNorm/RxClass identifiers (no model-invented drugs) and framed as clinical decision support with transparent reasoning — a human stays in the loop; the server makes no substitution decision.

The gap it fills: FDA/ASHP tell you a drug is short. Equivalence tools (Orange Book, vendor catalogs) tell you who else makes the same molecule. Neither tells you whether the different drug you'd actually reach for — a same-class alternative — is also in shortage. This stitches FDA shortage status + RxNorm/RxClass class alternatives + a per-alternative re-check into one LLM pass.

---

What it does

drug name ──▶ rx_normalize_drug ──▶ RxCUI + clean name
                                        │
                                        ▼
                              rx_get_drug_class ──▶ candidate ATC-4 classes
                                        │
                                        ▼
                            rx_find_alternatives ──▶ sibling drugs in the class
                                        │
                   ┌────────────────────┴───────────────────┐
                   ▼  (LLM loops over each sibling + original)
             rx_check_shortage ──▶ Current / Resolved / no record
                   │
                   ▼
        ranked candidates, each flagged if ALSO short   (the LLM synthesizes)

The fan-out loop is the model's reasoning, not a single tool call — that's why this is an MCP server, not a script.

The four tools

Tool	Purpose	Source
rx_normalize_drug	Messy name / brand / typo → RxCUI + ingredient	RxNorm (Prescribe)
rx_get_drug_class	Drug → candidate ATC-4 pharmacologic classes	RxClass
rx_find_alternatives	Class → sibling drugs (carries safety disclaimer)	RxClass
rx_check_shortage	Drug → current shortage status	openFDA Drug Shortages

All four are read-only. No API key required (see below).

Example

Ask your MCP client: "Furosemide is on backorder — what could be considered instead, and are any of those also short?" The model chains the four tools and synthesizes a ranked, shortage-flagged shortlist:

Same-class option (ATC C03CA)	FDA status	Presentations affected	Updated
furosemide (original)	Current shortage	33	06/25/2026
bumetanide	Current shortage	12	06/25/2026
torsemide	No shortage reported	0	—
piretanide	No shortage reported	0	—

The cascade insight: both first-line loop diuretics — furosemide and its usual substitute bumetanide — are in current FDA shortage; torsemide is the same-class option with no shortage reported. Candidates for a licensed professional to evaluate — not a substitution instruction; status ≠ availability, and class membership ≠ clinical interchangeability.

(Live national data — your results will reflect current shortage status, which changes over time.)

🎥 Demo video: coming soon.

Requirements

• uv (recommended) — handles Python + deps automatically
• Python ≥ 3.10 (uv fetches 3.12 for you if absent)
• An MCP client — Claude Desktop or Claude Code

Install & run

git clone https://github.com/Travis-Clement-Dev/rx-shortage-mcp.git
cd rx-shortage-mcp
uv run python -m rx_shortage_mcp   # first run auto-creates the venv + installs deps

The server speaks MCP over stdio; you normally launch it through an MCP client rather than by hand.

Claude Desktop

Add to claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "rx-shortage": {
      "command": "uv",
      "args": ["--directory", "/ABSOLUTE/PATH/TO/rx-shortage-mcp", "run", "python", "-m", "rx_shortage_mcp"]
    }
  }
}

Restart Claude Desktop, then ask: "Furosemide is on backorder — what could I consider instead, and are any of those also short?"

Optional: openFDA API key

The server runs fully keyless (~240 req/min, ~1000 req/day per IP). A free openFDA key only raises the daily ceiling. To use one, copy .env.example to .env and set OPENFDA_API_KEY.

Testing

uv run --extra dev pytest        # offline unit + safety-gate + MCP-protocol tests (deterministic)
uv run --extra dev pytest -m live # live eval: full chain over 15 drugs, asserts >=90% chain success
npx @modelcontextprotocol/inspector uv run python -m rx_shortage_mcp   # interactive tool testing

The safety gate (tests/test_safety_gate.py) fails the build if any tool ever frames output as a substitution instruction or drops the disclaimer.

Troubleshooting: if import rx_shortage_mcp ever fails, the editable install drifted — reset with rm -rf .venv uv.lock && uv sync. Let uv own the venv; don't mix in uv pip install -e ..

Safety

rx_find_alternatives returns every member of a pharmacologic class — some will be the wrong route, wrong indication, contraindicated, or withdrawn from market. The server cannot encode clinical judgment. Every alternatives response carries a disclaimer; output is framed as "candidate alternatives to consider," never "substitute with X." A licensed professional must evaluate every candidate. National-level shortage data only — no local/regional stock.

Attribution

This product uses publicly available data from the U.S. National Library of Medicine (NLM), National Institutes of Health, Department of Health and Human Services; NLM is not responsible for the product and does not endorse or recommend this or any other product.

Shortage data from the U.S. Food & Drug Administration (openFDA). openFDA data is not for making decisions regarding medical care.

License

MIT © 2026 Travis Clement