mcp-brasil-server

Pass

Audited by Gen Agent Trust Hub on May 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill includes tools such as tse_candidatos_query and inep_censo_query that allow the agent to execute arbitrary SQL queries against local DuckDB databases. This is an intended feature for data analysis but represents a dynamic execution surface.
  • [EXTERNAL_DOWNLOADS]: The skill downloads the mcp-brasil package from official registries and may download large datasets (100MB to 1.6GB) from Brazilian government sources (TSE, ANP, INEP, etc.) when specific environment variables are enabled.
  • [INDIRECT_PROMPT_INJECTION]: The skill connects to over 70 external Brazilian public APIs. This creates a significant ingestion surface where malicious or unexpected content in API responses could potentially influence agent behavior.
  • Ingestion points: Data retrieved from 70+ government and public APIs across health, economy, education, and transparency sectors.
  • Boundary markers: The instructions do not specify explicit delimiters or "ignore instructions" wrappers for the data returned by these APIs.
  • Capability inventory: The server can execute batch queries (executar_lote) and generate execution plans (planejar_consulta) based on user objectives.
  • Sanitization: While tool discovery uses BM25 filtering, there is no explicit mention of sanitizing or escaping the content of the API responses before they are processed by the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
May 16, 2026, 06:50 PM
Security Audit — agent-trust-hub — mcp-brasil-server