Strong naming conventions (verb_noun pattern consistently applied). All 17 tools have descriptions (avg ~120 chars, within baseline 34-392). Input schemas present with type definitions and constraints (minLength, minimum, maximum, pattern, enum). However, output schemas are not documented in the source, LLMs cannot predict response structure. Write operations require explicit 'confirmed': true, showing confirmation-request pattern adoption. French descriptions are clear but reduce discoverability for non-French-speaking agents. No error recovery guidance in descriptions.

• Document output schemas for all tools. For read operations (get_work, get_user, etc.), specify returned fields, types, and structure. Example: 'Returns {id: number, title: string, rating: number, reviews: [{id, author, text}], ...}'.
• Provide English descriptions alongside French. Use bilingual format: 'Search works on SensCritique / Rechercher des œuvres sur SensCritique.' This enables non-French agents to understand tool purpose.
• Add recovery guidance to descriptions of lookup tools. Example: 'If work not found, try senscritique_search_works() with a partial title first.' This guides multi-step planning.
• Document pagination response structure. Specify whether responses include 'total_count', 'has_more', or 'next_offset' so LLMs know how to iterate through large result sets.
• Clarify the confirmation-request pattern in write operation descriptions. Example: 'Requires explicit user confirmation (confirmed: true) to prevent accidental rating changes. Always ask the user before calling this tool.'