Server has 4 tools with basic schemas and descriptions in Portuguese. All tools have input schemas with type definitions and descriptions. However, descriptions are minimal (10-80 chars, below the 50-200 char baseline for LLM optimization), parameter descriptions lack actionable constraints (e.g., phone format, timeout bounds), and output schemas are completely undocumented. No error handling guidance, no idempotency hints, and no security/permission documentation. Tool naming follows verb_noun pattern (good), but lacks clarity on side effects and prerequisites.

• Document output schemas for all tools. Example: getWhatsAppQrCode should return {qr_code: string (base64 or URL), expires_at: ISO8601, instance_name: string}. This enables LLMs to chain calls and extract required fields.
• Expand tool descriptions to 50-200 characters with WHEN/WHY/WHAT structure. Example: 'Creates a new WhatsApp instance in Evolution API. Call this first to set up a new bot. Returns instance_name for use in other tools. Requires valid Evolution API credentials.'
• Add parameter constraints in descriptions. Example: phoneNumber: 'International format: +<country_code><area_code><number>, e.g. +5511987654321. No spaces or special characters.' timeout: 'Seconds to wait for QR code (1-300, default 60).'
• Implement error recovery guidance. Example: if phone validation fails, return 'Invalid phone format. Expected +<country><area><number>, got <value>. Try: +5511987654321'. If rate limited, return 'Rate limit exceeded. Retry after 60 seconds.'
• Add idempotency hints and side-effect warnings. Mark sendWhatsAppMessage as destructive in description: 'Sends a message immediately, this action cannot be undone. Verify phone and message content before calling.' Consider adding a dry_run parameter to preview without sending.