A leave management system MCP server for managing employee leave applications, balances, and history.
This server has 3 tools with mixed quality. Naming is clear and verb-based (apply_leave, get_leave_balance, get_leave_history). Descriptions exist but are very brief (10-47 chars), falling below the 34-char p10 baseline. Schemas are present with proper types and descriptions for most parameters. However, descriptions are too terse to guide LLM selection effectively, parameter descriptions lack detail on constraints and usage context, and there is no documentation of output schemas or error handling. The server lacks pagination support despite operating on history lists, lacks error recovery guidance, and stores data in-memory without persistence, which is not production-ready. Based on rubric calibration: descriptions are under 50 chars (should be 50-200 for LLM optimization), parameters lack constraint documentation, and error handling is minimal.
Apply for leave with dates, reason, and type. Example: - leave_dates = ["2025-05-10", "2025-05-11"] - reason = "Family function" - leave_type = "casual"
Check current leave balance for an employee.
Get detailed leave history for an employee.
Description too brief (10-47 chars vs 50-200 optimal). 'Check current leave balance for an employee.' and 'Get detailed leave history for an employee.' lack context for LLM selection and are at/below the p10 baseline of 34 chars. They do not explain when to use this tool vs others or what result to expect.
Output schemas not documented. apply_leave returns a string, get_leave_history returns List[dict], but there is no documented schema describing the structure of the dict or what fields appear in responses. LLMs cannot plan downstream calls without knowing the shape of returned data.
No pagination support for get_leave_history. If an employee has 100+ history entries, the entire list is returned, wasting tokens and risking context window exhaustion. No limit parameter, no offset/cursor, no total count. Agents cannot paginate large result sets.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 44 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 35 | - | v1 |
Error handling lacks recovery guidance. 'Employee ID not found.' and 'Insufficient leave balance.' are bare messages with no actionable next steps. No suggestion to search/list employees, no hint on how to free up balance. Pattern requires: 'User not found. Try search_users() with a partial name.'
Parameter 'leave_type' in apply_leave lacks enum constraints. Description says 'e.g., casual, sick, personal' but does not declare these as the only valid options. LLMs may hallucinate values like 'unpaid' or 'bereavement'. Use enum: ['casual', 'sick', 'personal'] or expand with valid types.
Parameter descriptions lack format/constraint details. 'leave_dates' says 'List of leave dates in string format' but does not specify ISO 8601, does not mention range validation, and does not warn against past dates. 'reason' has no length limit or content guidance.
No confirmation or dry-run for destructive operation (apply_leave). Agents can approve leave immediately without a confirmation step. Pattern requires: 'Irreversible operations (delete, send, publish) should support a dry-run or confirmation step.' This risks unintended approvals.
In-memory data store with no persistence. All leave records are lost when the server restarts. Not suitable for production. No audit trail or logging of who made changes and when.
Tool composition issue: apply_leave returns only a summary message, not the created leave record(s). If an agent needs to verify the leave was created or reference it in a followup call, it has no leave_id or transaction_id to work with. Should return structured object with leave IDs for chaining.