Healthcare Data MCP

Workflow recipes for source-first healthcare AI.

Start with the task, retrieve the evidence, preserve the receipt.

Browse data sources View repository
5public examples on page
11executable workflow plans
19local servers
v0.2.0release

Recipes, not vague prompts.

Each recipe starts with identifiers, names the workflow plan and servers, separates source families, and states which evidence receipt must travel into the final answer. The hard boundary is as important as the retrieval path.

Compliance Screen Health System Profile Market Scan Quality Lookup Facility Comparison Evidence Receipts

Task-first recipe cards.

These examples mirror the v0.2.0 workflow contract: public/reference data only, exact identifiers first, source caveats preserved, and no professional authority claims.

Compliance Screen

Best input identifiersNPI, entity name, CCN, state, UEI/CAGE where available.
Workflow IDscompliance_exclusion_screening
Recommended serverspublic-records, provider-enrollment, optional live-gateway
Source families
  • HHS OIG LEIE
  • SAM.gov exclusions
  • PECOS enrollment
Agent steps
  1. Check LEIE metadata/cache readiness.
  2. Use exact NPI screening when NPI exists.
  3. Use name search only as potential-match context.
  4. Use SAM.gov only when API key is configured.
  5. Preserve match basis, confidence, source metadata, and caveat.
Evidence to preserve
  • hhs_oig_leie receipt
  • SAM.gov source metadata when used
  • PECOS identity/enrollment receipt
  • Candidate-vs-exact match basis
Limits / do not claim
  • No legal clearance.
  • No final identity verification.
  • No exclusion conclusion without source-specific context.
Output exampleScreening-status table with one row per source, match basis, confidence, source URL/period, caveat, and unresolved follow-up questions.
hc-mcp workflow compliance_exclusion_screening --input npi=1234567890 --input entity_name="Example Health" --json

Health System Profile

Best input identifiersSystem name, AHRQ system ID if known, CCNs, NPIs, domain.
Workflow IDssystem_reconciliation, facility_profile
Recommended servershealth-system-profiler, cms-facility, provider-enrollment, web-intelligence, service-area, workforce-analytics
Source families
  • AHRQ Compendium
  • CMS POS
  • CMS hospital general info
  • PECOS and NPPES
  • Public web
Agent steps
  1. Resolve exact AHRQ/system/facility identifiers first.
  2. Treat web pages and names as aliases unless exact IDs agree.
  3. Attach facility-level receipts for system affiliation.
  4. Preserve identity-map conflicts.
Evidence to preserve
  • AHRQ system/facility linkage receipts
  • CCN/NPI identity maps
  • POS and hospital general info source metadata
  • Public-web alias evidence when used
Limits / do not claim
  • No common ownership from marketing names alone.
  • No affiliation from NPPES name-search hits alone.
  • No conflict hiding; unresolved identifiers stay visible.
Output exampleReconciled system/facility roster with identifiers, source-specific names, conflicts, and evidence per facility row.
hc-mcp workflow system_reconciliation --input query="Jefferson Health" --input system_slug=jefferson-health --json

Market Scan

Best input identifiersMarket geography, ZIP/ZCTA/county, CCNs, drive-time assumptions.
Workflow IDsmarket_community_health_scan
Recommended serversgeo-demographics, community-health, service-area, drive-time
Source families
  • Census ACS
  • CDC PLACES
  • CMS HSAF
  • Dartmouth HSA/HRR
  • CMS Geographic Variation
  • Routing backend
Agent steps
  1. Define geography before comparing facilities.
  2. Separate community indicators from facility outcomes.
  3. Preserve routing assumptions and source period.
  4. Label modeled access scores as modeled context.
Evidence to preserve
  • Geography identity and source period
  • CDC/Census receipts
  • HSAF/HSA/HRR receipts
  • Routing backend and assumptions
Limits / do not claim
  • No all-payer market share.
  • No network adequacy.
  • No patient leakage or facility quality from geography alone.
Output exampleMarket context memo separating geography definition, community indicators, access model assumptions, service-area evidence, and open verification questions.
hc-mcp workflow market_community_health_scan --input county_fips=42101 --input zcta=19104 --json

Quality Lookup

Best input identifiersCCN and exact measure ID.
Workflow IDsquality_measure_lookup
Recommended servershospital-quality, discovery
Source families
  • CMS Provider Data Catalog hospital quality files
Agent steps
  1. Use exact row-level measure lookup.
  2. Preserve rows[].evidence.
  3. Report when an exact measure is absent instead of substituting adjacent summaries.
Evidence to preserve
  • CCN
  • Exact measure ID
  • CMS source URL and source period
  • Row-level evidence receipt
Limits / do not substitute
  • HAC total for CLABSI SIR.
  • HRRP condition rows for hospital-wide readmission.
  • PHC4 mortality for CMS 30-day AMI mortality.
Output exampleMeasure lookup result showing exact row presence/absence, score fields when present, source period, and row evidence.
hc-mcp workflow quality_measure_lookup --input ccn=390223 --inputs-json '{"measure":"clabsi_sir"}' --json

Facility Comparison

Best input identifiersTwo or more CCNs.
Workflow IDsquality_profile, optional hospital_competitive_profile
Recommended servershospital-quality, cms-facility, optional claims-analytics, financial-intelligence, workforce-analytics
Source families
  • CMS quality
  • CMS hospital general info
  • POS
  • Cost reports
  • Claims PUFs where configured
Agent steps
  1. Anchor every row to CCN.
  2. Compare exact same measure IDs across facilities.
  3. Keep source periods visible.
  4. Separate quality, claims, finance, and workforce evidence.
Evidence to preserve
  • Facility identity receipts
  • Exact measure receipts
  • Cost report/claims/workforce source periods
  • Missing-data caveats by source
Limits / do not claim
  • No current operational superiority from stale source periods.
  • No facility comparison from mismatched measures.
  • No all-payer conclusion from Medicare public aggregates.
Output exampleComparison matrix with one row per CCN/measure/source family and a separate caveat column for source period and data availability.
hc-mcp workflow quality_profile --input ccn=390223 --input comparison_ccn=390111 --json

Evidence receipt examples.

An evidence receipt is the minimum citation unit an agent must preserve before turning retrieved data into a claim. These examples show the required fields and conservative caveats; they are example shapes, not live query results.

  • source_name
  • source_url
  • dataset_id
  • source_period
  • retrieved_at or source_modified
  • cache_status
  • cache_freshness
  • match_basis
  • confidence
  • caveat
  • agent_use

LEIE screening receipt

{
  "source_name": "HHS OIG LEIE",
  "source_url": "https://oig.hhs.gov/exclusions/downloadables/UPDATED.csv",
  "dataset_id": "hhs_oig_leie",
  "source_period": "current downloadable LEIE file",
  "retrieved_at": "copy_from_tool_evidence",
  "cache_status": "copy_from_tool_evidence",
  "cache_freshness": "31-day freshness target",
  "match_basis": "exact NPI when supplied; name search is potential-match context",
  "confidence": "copy_from_tool_evidence",
  "caveat": "Screening support, not legal clearance.",
  "agent_use": "Used as current public exclusion-screening evidence for the supplied NPI."
}

CMS quality measure receipt

{
  "source_name": "CMS Provider Data Catalog",
  "source_url": "copy_from_tool_evidence.source_url",
  "dataset_id": "cms_hospital_quality",
  "source_period": "copy_from_tool_evidence.source_period",
  "retrieved_at": "copy_from_tool_evidence.retrieved_at",
  "cache_status": "copy_from_tool_evidence.cache_status",
  "cache_freshness": "copy_from_tool_evidence.cache_freshness",
  "match_basis": "exact CCN plus exact measure ID",
  "confidence": "source row when exact_measure_found=true",
  "caveat": "Adjacent summaries are not substitutes for exact measure rows.",
  "agent_use": "Used exact CCN + measure row as support for a quality fact."
}

Market/community receipt

{
  "source_name": "CDC PLACES or Census ACS",
  "source_url": "copy_from_tool_evidence.source_url",
  "dataset_id": "cdc_places",
  "source_period": "copy_from_tool_evidence.source_period",
  "retrieved_at": "copy_from_tool_evidence.retrieved_at",
  "cache_status": "copy_from_tool_evidence.cache_status",
  "cache_freshness": "live/API-backed or cache metadata",
  "match_basis": "geography code such as county, tract, ZCTA, or place",
  "confidence": "source geography match; not facility-specific",
  "caveat": "Not facility-specific performance and not patient-level evidence.",
  "agent_use": "Used as community/market context for a geography."
}