Six tools with clear verb-noun naming (list, transcript, outline, detail, folders, search) and reasonable descriptions (60-90 chars). All tools have input schemas with typed parameters and descriptions. However, output schemas are completely undocumented, LLMs cannot predict what fields to expect from responses. No error handling guidance, no pagination support despite list/search potentially returning many results, and no indication of which tools are read-only vs. destructive (all are read-only, but this should be explicit). Parameter descriptions lack format constraints (e.g., date format for 'after'/'before' is mentioned but not formalized). Missing idempotency hints and recovery guidance.

• Document output schemas for all tools. E.g., 'list' returns {files: [{id, filename, duration, start_time, folders}], total_count}. 'transcript' returns {segments: [{speaker, content, start_time}]}.
• Add pagination to 'list' and 'search': accept limit (1-100, default 20) and offset/cursor parameters; return total_count and next_cursor in response.
• Formalize date constraints: add 'pattern': '^\d{4}-\d{2}-\d{2}$' to 'after' and 'before' parameters in schema.
• Add error recovery hints to descriptions: 'If recording_id is invalid, call list() first to find valid IDs. If transcript is still processing, try again in 30 seconds.'
• Declare tool annotations in schema: add 'readOnlyHint': true to all tools to signal they are safe to retry and do not modify state.