MCP server for Romanian public sector salary calculations based on three official bill variants (May 25, July 17, August 20, 2026). Provides tools to search job functions, calculate salaries with tax deductions, list annexes, and retrieve law articles.
Strong schema definitions with Zod validation and comprehensive parameter descriptions. All 5 tools have explicit schemas with type constraints and detailed Romanian-language documentation. Tool names follow verb_noun pattern (search_function, calculate_salary, list_anexe, list_variante, get_law_article). However, output schemas are not formally documented in the code, only input schemas are visible. Error handling is minimal; no recovery guidance or actionable error messages. Descriptions are domain-specific but lengthy (200+ chars), which is acceptable for complex financial/legal tools but could be more concise.
Calculează salariul brut și net pentru o funcție, cu gradații de vechime, sporuri și impozitare (CAS/CASS/impozit). Parametrul opțional `varianta` alege textul proiectului (implicit 20 august 2026, VR 4.000 lei).
Rezumat al articolelor cheie din proiectul de lege (1-41), pe varianta aleasă.
Listează cele 9 familii ocupaționale (anexele I-IX) cu numărul de funcții din fiecare.
Cele trei variante oficiale ale proiectului (25 mai, 17 iulie, 20 august 2026): valoare de referință, intrare în vigoare, solde de grad, ce s-a schimbat și ce nu este modelat.
Caută funcția în datasetul variantei alese (2.627 / 2.809 / 3.000 de funcții — 25 mai / 17 iulie / 20 august 2026). Returnează coeficientul, anexa, gradul, studiile cerute.
Output schemas not documented. Tool implementations return structured objects but no formal schema definitions (TypeScript interfaces or JSON Schema) are visible for responses. LLMs cannot plan downstream calls without knowing return field names and types.
No error handling or recovery guidance. Tools do not document failure modes, validation errors, or what the LLM should do if a call fails (e.g., 'varianta not found' or 'coeficient out of range'). Errors will not guide agent recovery.
Parameter descriptions are lengthy (200+ chars) and mix multiple concerns. E.g., calculate_salary.coefIncludeVechime description spans 300+ chars explaining law articles and edge cases. Descriptions should be 50 - 150 chars; move legal details to separate documentation.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | 2026-07-28+ | v2 |
No pagination or result limits documented for search_function. The tool accepts a 'limit' parameter (max 50) but does not document total count, next_cursor, or how to iterate large result sets. LLMs may not know how to fetch all results.
Tool annotations (readOnlyHint, destructiveHint, idempotentHint) are absent. All tools are read-only (no state modification), but this is not declared in the tool definitions. Modern MCP servers should annotate tool safety properties.