Loading ToolBench
Exposes 5 tools for safe patient care transitions using Po FHIR data and Groq AI. Generates structured handoff documents, identifies follow-up gaps, performs medication reconciliation, and assesses readmission risk.
Care Transition MCP has well-intentioned tool design but significant definition gaps. All 5 tools have descriptions (good), but 4 of 5 have EMPTY input schemas (critical failure). Tool names follow verb_noun convention (get_, generate_, flag_, medication_, readmission_). However, descriptions lack actionable detail on when to use each tool vs. alternatives, and parameter descriptions are missing entirely. The generate_transition_brief_tool is the only tool with a documented parameter (transition_type with enum-like description), but even that lacks formal schema constraints. Error handling is present (returns error dict when patient context missing) but not recovery-focused. No output schemas documented. This is a D-grade server, functional but not production-ready for agent use.
Identify follow-up care gaps before or after a patient transition. Flags missing appointments, abnormal labs, and medication follow-up needs. Patient context is injected automatically by Po when a patient is selected.
Generate a structured care transition brief for a patient being discharged. Uses FHIR data and Groq AI to produce a clinician-ready handoff document.
Get a structured summary of a patient's clinical profile. Returns name, date of birth, active conditions, and current medications. Patient context is injected automatically by Po when a patient is selected.
Check for medication discrepancies in a patient's current medication list. Flags duplicates, missing medications for known conditions, and dosage concerns. Patient context is injected automatically by Po when a patient is selected.
Generate a plain-language readmission risk narrative for a patient. Returns a risk level (LOW/MODERATE/HIGH), key risk drivers, and recommended actions for the care team. Patient context is injected automatically by Po when a patient is selected.
4 of 5 tools have empty input schemas ({}). This prevents LLMs from understanding parameter constraints and inviting hallucinated values.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., does get_patient_summary return patient_id for chaining?). Violates pattern:tool and mxe:include-chaining-ids.
Tool descriptions lack WHEN-to-use guidance and differentiation. E.g., flag_followup_gaps_tool and medication_reconciliation_tool both operate on patient context but descriptions don't explain when to call one vs. the other. Violates pattern:tool-description.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
generate_transition_brief_tool has a transition_type parameter but no enum constraint in schema. Description mentions 'hospital-to-home, ICU-to-ward, hospital-to-SNF, ED-to-admission' as examples, not formal constraints. LLMs will hallucinate invalid values.
Error handling returns generic 'No patient selected' message. Does not guide recovery (e.g., 'Select a patient in Po first, then retry') or classify error as retryable. Violates pattern:recovery-guide.