MCP server modeling staffing and compliance agents over a synthetic home-care workforce dataset.
Four tools with basic schemas and descriptions, but significant gaps in parameter documentation and error guidance. All tools have descriptions (10-200 chars), and input schemas are present with type definitions. However, parameter descriptions are minimal or missing context. Error handling is present but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint). Output schemas are implicit (model_dump() calls) rather than explicitly documented. Naming is clear and verb-driven (list_, suggest_, assign_). Composition is reasonable, each tool has a single responsibility, though suggest_assignments and assign_shift could benefit from clearer chaining documentation.
Here I assign a caregiver to a shift if valid.
Here I list caregivers with compliance items expiring soon.
Here I list open shifts in a time window.
Here I suggest caregiver assignments for open shifts. Simple heuristic: - role and skills must match - caregiver must belong to location - prefer caregivers whose preferred_shift_types match the shift time
Parameter descriptions lack actionable constraints. 'location_id' described as 'Optional location filter (e.g. 'loc_nyc')', example values invite literal reuse. Should specify format pattern and valid location IDs.
Error responses lack recovery guidance. assign_shift returns {'ok': False, 'error': 'shift_not_found'} with no hint to call list_open_shifts() first. Errors should guide the LLM to the next action.
Output schemas not explicitly documented. Tools return model_dump() dicts, but LLMs cannot see the Shift/Caregiver/ComplianceItem schema structure. Downstream tools (e.g., assign_shift) expect shift_id, but list_open_shifts response structure is implicit.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
No tool annotations (readOnlyHint/destructiveHint). assign_shift is a WRITE operation but lacks destructiveHint=true. list_* tools lack readOnlyHint=true. Annotations help LLMs reason about side effects and retry safety.
suggest_assignments strategy parameter lacks enum constraint. Accepts 'fair_load' but no validation or list of valid strategies. Should declare enum: ['fair_load', 'skill_match', 'availability_first'] or similar.