Model Context Protocol server for Brazilian public data lookup (CNPJ companies and CEP postal codes) with intelligent web search and sequential thinking
Three tools with variable quality. All have descriptions and explicit schemas, but lack depth in parameter documentation and output schema specification. Tool naming follows verb_noun convention (lookup_*, search) which is good. However, descriptions are minimal (10-60 chars each), parameter descriptions are sparse or missing, and output schemas are not documented. Error handling exists but lacks actionable recovery guidance. No tool annotations (readOnlyHint, etc.) despite all being read-only operations.
Look up CEP postal code information from Brazilian public databases
Look up CNPJ company information from Brazilian public databases
Perform intelligent web search using Tavily API with optional reasoning
Parameter descriptions are minimal or missing. 'cnpj' and 'cep' params have trivial descriptions ('CNPJ number to look up', 'CEP postal code to look up'). The 'reasoning' boolean parameter in search tool lacks any description of what enabling it does or when to use it. LLMs cannot infer parameter meaning from names alone.
Tool descriptions are below baseline length (10-60 chars vs recommended 50-200). 'Look up CNPJ company information from Brazilian public databases' (59 chars) lacks context for LLM selection: What format is returned? Are there limitations? When should this be called vs search? Incomplete descriptions force LLMs to guess.
Output schemas are not documented. The mcp-server.ts code shows tools return LookupResult with 'ok' and 'data' fields, but the schema for 'data' is not defined. Agents cannot plan downstream calls or parse responses correctly without knowing field names and types.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 45 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 44 | - | v1 |
search tool has ambiguous naming. The generic name 'search' does not distinguish it from other search tools or clarify that it uses Tavily API for web search. A name like 'search_web' or 'tavily_search' would be clearer. Current name risks LLM confusion with local data lookup tools.
No tool annotations present. All three tools are read-only (marked Risk: READ_ONLY) but lack readOnlyHint annotations in the schema. This prevents agents and clients from understanding that these tools cannot modify state and are always safe to call. Current spec (2026-07-28) expects tool annotations.
Error messages lack recovery guidance. The formatErrorResponse() function in mcp-server.ts returns error code and message but does not suggest next steps. When lookup_cnpj fails (e.g., invalid CNPJ), the agent receives only an error code, not 'Try search() to find the company name first' or 'CNPJ must be 14 digits'.
search tool 'reasoning' parameter lacks constraint documentation. Is reasoning always applicable? Does it change cost, latency, or output format? LLMs need explicit guidance on when to enable it. Current description says nothing about these trade-offs.