Fraud Detection (Anthropic Healthcare Plugin)

Anthropic-published skill from the anthropics/healthcare plugin marketplace that screens Medicare and Medicaid claims for fraud, waste, and abuse and produces ranked investigation referrals with full citations.

   
Type Claude Skill (shipped inside the healthcare Claude Code plugin)
Supplier Anthropic
Availability GA
Pricing Free / OSS — provided under Anthropic’s terms of service
Capabilities Read/Write — reads a claims corpus, writes referral packets and a dashboard
Verified works · 2026-07-20
Security caution · 2026-07-20 — first-party Anthropic, skill dir confirmed, but repo has no LICENSE and it handles claims data

How to install

  • Claude Code — plugin marketplace: 
    /plugin marketplace add anthropics/healthcare
/plugin install healthcare@healthcare

    The fraud-detection skill is bundled inside the consolidated healthcare plugin (the older standalone plugins are now deprecated in favor of healthcare@healthcare). Invoke it as /healthcare:fraud-detection (namespaced by the plugin — not a bare /fraud-detection).

  • Claude.ai / Claude Desktop — the healthcare plugin’s skills load wherever the marketplace plugin is enabled; the enrichment MCPs (ICD-10, NPI, CMS Coverage) are optional and connect over the hosted hcls.mcp.claude.com endpoints.

What it does

Runs a three-tier claims-screening pipeline:

  • Deterministic detection — rule-based screening of a claims corpus against public rulesets (NCCI MUE, OIG LEIE exclusions, CMS enrollment, Physician Fee Schedule).
  • Model adjudication — the model narrates patterns and may dismiss or downgrade findings but never adds new allegations.
  • Synthesis — generates investigator narratives with adversarial verification.

Input is a claims corpus in a DuckDB canonical six-table schema plus a quarter and line of business (Medicare / Medicaid); reference rulesets are fetched from CMS, OIG, and NLM and cached locally. Output is ranked referrals with per-provider HTML packets, an index.html dashboard, and an Excel export. A “citation-or-zero” gate requires every dollar and rule allegation to trace to a deterministic recompute, and findings are framed as “indicators consistent with [scheme]” rather than definitive fraud.

Primary use cases: Special Investigation Unit (SIU) referral generation, program-integrity screening, auditable claims review.

Notes

Pairs with the icd10-cm, procedure-coding, and prior-auth skills and the ICD-10 / NPI Registry / CMS Coverage MCPs in the same healthcare plugin. Output is decision-support, not a legal determination — establishing intent is a downstream determination. Requires the caller to supply a claims corpus in the expected DuckDB schema.

Sources

Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.