posted an update

Prompt Opinion integration: FHIR context trusted, first tool live

Quick milestone for DaktariTB.

The MCP server is deployed, registered with Prompt Opinion, and the FHIR context extension (ai.promptopinion/fhir-context) is trusted. The first clinical tool — order_tb_workup — is live and has been invoked end-to-end on a test patient.

The test case

A fictional 34-year-old PLHIV patient in Nairobi on TLD (dolutegravir-based ART), presenting with:

  • 4-week productive cough
  • 8 kg unintentional weight loss
  • Drenching night sweats × 3 weeks
  • CD4 count dropped from 480 → 290 cells/μL over six months

What the agent did

Working inside a Prompt Opinion chat with DaktariTB MCP wired in:

  1. Fetched the patient's FHIR bundle via SHARP context headers
  2. Recognized the WHO 4-symptom positive screen
  3. Added Urine LF-LAM to the workup because CD4 < 350 (per WHO guidance for PLHIV)
  4. Created 4 FHIR ServiceRequest resources, each linked to her HIV Condition via reasonReference:
    • GeneXpert MTB/RIF (LOINC 88142-3)
    • AFB sputum smear microscopy (LOINC 648-0)
    • Chest X-ray (LOINC 36554-4)
    • Urine LF-LAM (LOINC 95745-4)
  5. Returned structured output with fhir_id per order, orders_created: [...], errors: []

All four ServiceRequests verified persistent in the FHIR store via direct query.

What's next

  • adjust_art_for_rif — dolutegravir ↔ rifampicin UGT1A1 interaction; writes a DetectedIssue + supplementary DTG 50 mg BID MedicationRequest
  • generate_tb_notification — Kenya MOH TB-001 notification form as a FHIR DocumentReference

Built on MCP + FHIR R4 + Kenya MOH clinical protocols. Source discipline: open standards, no vendor lock-in, every action is an auditable FHIR resource.

Log in or sign up for Devpost to join the conversation.