Inspiration
Prior authorization is one of the most painful administrative workflows in healthcare. Before patients can receive many expensive medications or procedures, clinicians have to prove medical necessity to the payer. That usually means gathering patient history, labs, medication trials, payer rules, denial reasons, and appeal documentation by hand.
PriorPilot was built to show how Prompt Opinion, MCP, and FHIR context propagation can turn that fragmented workflow into reusable healthcare AI infrastructure.
What it does
PriorPilot is an AI prior authorization and appeal copilot integrated into the Prompt Opinion platform as an MCP server.
It exposes specialized tools that a Prompt Opinion agent can invoke to:
- summarize patient context
- extract clinical evidence
- evaluate mocked payer criteria
- generate a prior authorization packet
- generate an appeal letter after denial
- show an A2A-style workflow trace
The demo workflow uses a synthetic patient with Type 2 diabetes, obesity, elevated HbA1c, and documented metformin failure. PriorPilot uses that evidence to generate an Ozempic prior authorization packet and denial appeal.
How we built it
PriorPilot is implemented as a Python MCP server using FastAPI and the Prompt Opinion community MCP template.
The server exposes these MCP tools:
PriorPilotPatientSummaryPriorPilotWorkflowTracePriorPilotExtractEvidencePriorPilotEvaluatePolicyPriorPilotGeneratePacketPriorPilotGenerateAppeal
Prompt Opinion discovers and invokes these tools through a Streamable HTTP MCP endpoint. The server supports Prompt Opinion FHIR context headers, including patient ID and FHIR server URL, so tools can be grounded in active patient context. For demo reliability and safety, the project uses synthetic data only.
SHARP / FHIR context support
PriorPilot supports Prompt Opinion’s healthcare context propagation model. The MCP server advertises FHIR context support and can read context headers such as:
x-fhir-server-urlx-fhir-access-tokenx-patient-idThis allows the workflow to understand which patient and FHIR environment are active. In the demo, PriorPilot uses a deterministic synthetic patient fallback so the workflow is reproducible and contains no PHI.
Why it matters
Prior authorization creates delays, staff burden, and payer-provider friction. PriorPilot shows how AI can help at the last mile: not just answering questions, but producing operational deliverables that clinicians and staff actually need.
The important idea is that PriorPilot is not another chatbot. It is interoperable workflow infrastructure. Each MCP tool is reusable and can be composed into other healthcare agents inside Prompt Opinion.
Challenges we ran into
The main challenge was balancing realism with hackathon reliability. Real payer integrations and live EHR data are complex and risky for a short build. We solved this by mocking payer policy logic while keeping the workflow grounded in FHIR-style patient resources and Prompt Opinion context propagation.
Accomplishments
- Built a working MCP server integrated with Prompt Opinion
- Published discoverable prior authorization tools
- Demonstrated FHIR context propagation
- Built a full denial-to-appeal workflow
- Used only synthetic data, with no real PHI
- Produced a demo that runs inside the Prompt Opinion platform
## What’s next
Next steps would include connecting to richer FHIR datasets, expanding payer policy libraries, adding PDF/form generation, supporting additional therapies and procedures, and integrating staff task creation for follow-up workflows.
Log in or sign up for Devpost to join the conversation.