ChatKazi MCP defines 7 tools with complete input schemas and descriptions. All tools follow verb_noun naming (list_, start_, get_, send_). Descriptions are present and actionable (avg ~80 chars). However, output schemas are entirely undocumented, callers cannot predict response structure. Parameter descriptions lack format constraints (e.g., phone number format is mentioned but not enforced via pattern). No error handling guidance, no recovery hints, and no idempotency markers. Security concern: sessionId defaults to 'default' silently, risking unintended session reuse. Missing tool annotations (readOnlyHint, destructiveHint) despite clear read/write semantics.

• Document output schemas for all tools. Example: list_sessions should return {type: 'object', properties: {sessions: {type: 'array', items: {type: 'object', properties: {sessionId: {type: 'string'}, status: {type: 'string'}}}}}}. This enables LLMs to chain tools correctly.
• Add tool annotations: mark logout_session, send_whatsapp_text, send_whatsapp_media with destructiveHint=true; mark list_sessions, get_session_status, get_session_qr with readOnlyHint=true. This signals intent without requiring description parsing.
• Add phone number validation: use JSON Schema pattern '^[0-9]{10,15}$' for 'to' parameter in send_whatsapp_text and send_whatsapp_media. Include in description: 'Phone number in E.164 format without + or spaces (10-15 digits, e.g., 254712345678).'
• Enhance error responses with recovery hints. Example: if sessionId not found, return 'Session "xyz" not found. Available sessions: [list]. Call list_sessions() to discover valid session IDs.' This guides LLM self-correction.
• Make sessionId required in send_whatsapp_text and send_whatsapp_media, or document the 'default' session behavior explicitly. If 'default' is a fallback, state: 'If omitted, sends from the default session. Call list_sessions() to see available sessions.'