Loading ToolBench
MCP server for querying Brazilian judicial processes (TJES) via DataJud/CNJ and jurisprudence search
Three tools with complete input schemas and descriptions, but significant gaps in parameter documentation and output schema specification. Tool names follow verb_noun convention (consultar_, buscar_, inteiro_). Descriptions are present (avg 120 chars) but lack actionable context for LLM selection. No parameter descriptions beyond the bare minimum. Output schemas are completely undocumented, LLMs cannot predict response structure. Error handling is absent from tool definitions. No tool annotations (readOnlyHint, etc.) despite all being read-only operations.
Busca jurisprudencia do TJES por texto livre. Retorna processos com a respectiva ementa.
Consulta um processo do TJES no DataJud/CNJ: classe, orgao julgador, andamentos e lista de decisoes disponiveis.
Retorna o texto integral de todas as sentencas, decisoes e acordaos de um processo do TJES.
No output schema documentation. LLMs cannot predict response structure (fields, types, nesting). All three tools return unstructured text blocks instead of structured JSON.
Parameter descriptions are minimal or missing. 'numero' and 'termo' lack format guidance (e.g., 'with or without punctuation' is in tool description, not param). 'base' enum is documented in description but not in schema enum field.
No error handling guidance in tool definitions. No indication of what happens on invalid process number, network failure, or empty results. LLM cannot plan recovery.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | <=2025-11-25 | v2 |
No tool annotations despite all being read-only. Missing readOnlyHint=true would signal to LLMs that these are safe to call without confirmation.
buscar_jurisprudencia returns up to 15 results by default but no pagination support. Large result sets will be truncated without indication of total count or next_cursor.