Case study 05
Digital Health MCP Server
Giving AI agents a governed, standards-compliant way to reach health data — instead of letting them guess.
Problem
The interesting question in clinical AI is not whether a model can reason. It is what the model is allowed to see, where that data came from, and whether the codes it reads mean what it assumes they mean.
Pointing a language model at a health database directly fails all three tests at once. There is no boundary, no terminology grounding, and no auditable account of what was accessed.
Approach
I built a Model Context Protocol server that exposes health data and terminology services to AI agents through an explicit, typed interface.
- Designed and deployed a Digital Health MCP Server as the boundary between AI agents and health data — a named contract rather than open database access.
- Exposed terminology services alongside the data, so an agent can resolve what a code means instead of inferring it.
- Kept the underlying data standards-compliant (FHIR, SNOMED CT, ICD-11), which is what makes the responses interpretable in the first place.
- Enabled AI-assisted clinical and programmatic decision support on top of that interface.
Technology
- Model Context Protocol
- HL7 FHIR
- SNOMED CT
- ICD-11
- Terminology services
- Generative AI
- PostgreSQL
Impact
AI agents can now support clinical and programmatic decisions over health data that is governed, standards-compliant, and reached through an explicit boundary — which is the difference between a demo and something that can be operated.
- AI-assisted clinical and programmatic decision support over standards-compliant data.
- An explicit, reviewable boundary between AI agents and health data.
- Terminology resolution available to agents as a first-class service.
- A reusable pattern for bringing AI to regulated health data.