Server has 6 tools with complete Zod schemas and basic descriptions. However, descriptions are minimal (10-50 chars), parameter descriptions lack detail on constraints and formats, and error handling returns generic Portuguese text without recovery guidance. Tool names use Portuguese verbs (buscar, detalhe, carregar, analisar, preparar, monitorar) which are action-oriented but non-standard for English-speaking LLM ecosystems. No output schemas documented. No tool annotations (readOnlyHint, destructiveHint). Error messages do not guide LLM recovery. Composition is reasonable, tools chain logically (search → detail → analyze → prepare), but lack of structured output and missing parameter descriptions significantly reduce LLM usability.

• Document output schemas for each tool. E.g., buscar_vagas returns {jobs: [{id, title, company, location, ...}], total, hasMore}. Include all fields LLM needs for downstream calls.
• Expand parameter descriptions with format and constraint details. E.g., 'caminhoPerfil: Absolute path to JSON profile file (must exist, valid UTF-8, max 10MB)' instead of just 'Caminho do JSON do perfil'.
• Replace generic error messages with actionable recovery guidance. E.g., 'Profile file not found at /path/to/file. Verify path exists and is readable. Use carregar_perfil to validate first.' instead of 'carregar_perfil: erro desconhecido'.
• Add tool annotations via MCP SDK. Mark all tools with readOnlyHint=true since they do not modify state. This signals to agents that retries are safe.
• Translate tool names to English (search_jobs, get_job_details, load_profile, analyze_fit, prepare_application, monitor_deadlines) for consistency with LLM training data and broader ecosystem compatibility.