Server has 9 tools with complete Zod schemas and descriptions. Naming follows verb_noun convention (list_, create_, get_, send_). Descriptions are present (avg ~100 chars) but lack actionable context for LLM selection. Parameters are typed and constrained (regex patterns for phone numbers, URL format validation). Critical gaps: no output schema documentation, no error recovery guidance, no pagination support for list_sessions, no idempotency hints for write operations, and descriptions do not explain when to use each tool vs alternatives.

• Document output schema for each tool. Example for send_text_message: 'Returns {status: number, data: {messageId: string, timestamp: ISO8601, chatId: string}}'. This enables LLMs to chain tools and extract required fields.
• Enhance descriptions with dependency hints and selection criteria. Example: 'Get the QR code to authenticate a session. Call this after create_session returns SCAN_QR_CODE status. Alternative: request_pairing_code for phone-based auth without scanning.'
• Wrap API responses with structured error messages. Instead of raw HTTP status, return: '{isError: true, message: "Session not found. Create a new session with create_session first.", retryable: true}'. This guides LLM recovery.
• Add pagination to list_sessions: accept optional 'limit' (default 20, max 100) and 'offset' parameters. Return {sessions: [...], total: number, hasMore: boolean}. Prevents context window exhaustion.
• Add idempotency hints to write tools. Document: 'This operation is idempotent, calling with the same session and chatId will not send duplicate messages.' Add optional 'idempotencyKey' parameter for explicit deduplication.