This is a Python SDK client, not an MCP server. No MCP protocol implementation detected. Tool definitions exist in source code but lack MCP registration, schemas are minimal (only port_id/port_ids strings), and descriptions are in Portuguese without LLM-optimized English. The three tools are READ_ONLY API wrappers with no error recovery guidance, no pagination support, and no structured output documentation. Naming follows verb_noun convention (get_port_risk, get_ports_risk_async) but descriptions are brief (under 100 chars) and lack context on when to use each variant. No input validation, no permission gates, no audit trails, this is a thin HTTP client wrapper, not a production-grade agent tool.

• Wrap this SDK in an MCP server implementation (e.g., using mcp.server.Server) that registers these three tools with proper MCP tool definitions, schemas, and descriptions.
• Translate all descriptions to English and expand to 150 - 250 characters, explaining WHEN to use sync vs async vs batch: 'Fetch port congestion risk synchronously. Use for single-port queries. For multiple ports, prefer get_ports_risk_async to avoid sequential latency.'
• Add input schema constraints: port_id as a string matching regex ^[A-Z]{2}[A-Z0-9]{3}$ (UN/LOCODE format), with description 'UN/LOCODE port code (e.g., BRSSZ, CNSHA, NLRTM)'. For port_ids array, add minItems: 1, maxItems: 100.
• Document the PortRisk output schema in each tool definition: fields port_id, port_name, country, congestion_score (0.0 - 1.0), eta_delay_days, waiting_vessels (int), freight_volatility_index, updated_at (ISO 8601).
• Add error handling with recovery guidance: catch requests.HTTPError and return 'Port not found. Verify UN/LOCODE format (e.g., BRSSZ). Call with a valid code or search available ports.' For ImportError on async, return 'Async requires httpx. Install: pip install aetherx-oracle[async].'