MCP server for processing handwritten clinical triage notes using OCR and deterministic clinical rules, generating FHIR bundles for Irish healthcare systems
Single tool with adequate naming but significant gaps in schema documentation and parameter descriptions. Tool name 'ingest_handwritten_note' follows verb_noun convention and clearly conveys action. However, the input schema is minimal (only 'image_bytes' with basic type), lacks validation constraints, and the output schema is entirely undocumented. Description is present (194 chars) but lacks actionable guidance on when to use the tool, prerequisites, or error recovery. No parameter-level descriptions visible in schema. Tool performs clinical triage, a safety-critical operation, yet lacks error handling guidance, confirmation patterns, or audit trail documentation.
Processes raw visual capture of a clinical triage note. - Extracts context into variables without using inference for clinical decisions. - Utilises deterministic equations for category allocation. - Generates corresponding FHIR schemas bundle.
Output schema completely undocumented. Tool returns dict with fhir_bundle, assigned_triage_category, requires_human_approval, message, but no schema definition visible. LLMs cannot plan downstream operations or validate response structure.
Input parameter 'image_bytes' lacks description in schema. No guidance on expected image format, size limits, resolution, or encoding.
No error handling guidance. Tool processes clinical data and returns 'requires_human_approval: true' but provides no recovery path if extraction fails, OCR confidence is low, or Gemini parsing errors occur. Clinical tools must guide error recovery.
No confirmation or dry-run pattern for safety-critical clinical triage. Tool assigns MTS categories that determine patient priority, irreversible decisions should support confirmation before execution.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | <=2025-11-25 | v2 |
Tool description lacks dependency hints and prerequisites. Does not explain that Gemini multimodal extraction is used, that GCP credentials are required, or that Document AI is optional. LLMs cannot infer setup requirements.