WhatsApp MCP server has 12 tools with complete input schemas and descriptions, but quality is inconsistent. All tools have verb-noun naming (search_, list_, get_, send_, download_) and descriptions (avg ~80 chars). However, descriptions lack LLM-optimized guidance on WHEN to use each tool vs. alternatives (e.g., list_messages vs. get_message_context). Output schemas are not documented, LLMs cannot plan downstream calls. Error handling is absent from descriptions. Parameter descriptions are present but generic; many lack format/constraint guidance (e.g., 'phone number with country code but no + or other symbols' is buried in send_message description, not in recipient param). No tool annotations (readOnlyHint/destructiveHint) despite clear read/write distinction. Security concern: send_message and send_file accept recipient as free-form string, no validation guidance for LLMs on format. Pagination is present (limit, page) but no total_count or next_cursor documented.

• Document output schemas for all 12 tools. Specify field names, types, and which fields are required. Example: search_contacts returns [{contact_id: string, name: string, phone: string, last_message_time: ISO8601}].
• Add tool annotations: mark send_message, send_file, send_audio_message with destructiveHint=true; mark all read-only tools with readOnlyHint=true.
• Move recipient format constraints into the recipient parameter description: 'Phone number with country code (no + or symbols, e.g., 14155552671) or JID (e.g., [email protected] or [email protected] for groups).'
• Add error handling guidance to send_* tools: 'If recipient is invalid, returns error with format hint. If not connected, returns error suggesting reconnection. Retryable on network timeout.'
• Document pagination behavior: 'Returns up to limit results. If result count == limit, more results may exist; increment page and retry. No total_count available; iterate until result count < limit.'