Server has 12 tools with complete input schemas and descriptions present for all tools and parameters. Naming follows verb_noun convention (search_, list_, get_, send_, download_). However, descriptions are generic and lack LLM-optimization guidance. Output schemas are not documented. Error handling is absent from tool definitions. No tool annotations (readOnlyHint/destructiveHint) despite clear risk stratification (8 READ_ONLY, 4 WRITE). Parameter descriptions lack format constraints, ranges, and dependency documentation. No confirmation/dry-run pattern for destructive operations (send_message, send_file, send_audio_message, download_media). Missing guidance on when to use similar tools (e.g., get_chat vs get_direct_chat_by_contact).

• Add tool annotations: mark all READ_ONLY tools with readOnlyHint=true; mark send_* and download_media with destructiveHint=true. This enables LLMs to reason about side effects.
• Document output schemas for all tools. E.g., list_messages returns [{message_id, chat_jid, sender_phone_number, text, timestamp}]. Include all fields needed for downstream tool calls.
• Add confirmation/dry-run pattern to send_message, send_file, send_audio_message. E.g., add a 'dry_run' boolean parameter that returns 'Would send to {recipient}: {message}' without actually sending.
• Add numeric constraints to parameters: limit (1-100), page (0+), context_before/after (0-50). Document recipient format as regex: '^[0-9]{1,15}(@s\.whatsapp\.net|@g\.us)$' or phone number without +.
• Add disambiguation guidance in descriptions. E.g., get_chat: 'Use when you have the chat JID. For phone number lookup, use get_direct_chat_by_contact instead.'