15 tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. Most tools have 1-3 parameters with type info, but descriptions are generic (10-50 chars) and lack actionable guidance. No error handling patterns, no output schemas documented, no parameter constraints (enums, ranges). Tool names follow verb_noun convention (get_, set_, start_, record_, end_, add_, create_, import_, update_) which is good, but descriptions are too brief to guide LLM selection. Example: 'Get the current practice state...' (48 chars) lacks context on when to call vs alternatives. No tool has documented return structure. Composition is reasonable (single responsibility per tool), but chaining IDs are not explicitly returned in descriptions.

• Document output schemas for all tools. For get_state, specify: {learnerProfile: {name, level, ...}, sessions: [{id, date, score, ...}], questions: [{id, part, topic, text, ...}], ...}. Use JSON Schema format.
• Add parameter constraints: set_learner_name displayName should be 1-100 chars; create_practice_plan lengthDays should be 1-365; weeklyTarget should be 1-7. Use enum for 'part' (Part 1|Part 2|Part 3) and 'focus' (fluency|vocabulary|pronunciation|grammar).
• Add error recovery hints: 'If questionId not found, call get_questions() to list available IDs.' Include in tool descriptions or as separate error_guidance field.
• Expand tool descriptions to 100-200 chars with context: 'Start a new practice session with a selected question. Call get_questions() first to find a valid questionId. Returns sessionId for use in record_answer and end_session.'
• Clarify pagination: 'Returns up to 20 questions by default. Use limit (1-100) and offset for pagination. Total count not returned; iterate until fewer than limit results.' Add total_count or next_cursor to response schema.