37 tools with consistent verb_noun naming (transport_*, track_*, fx_*, etc.) and explicit JsonSchema parameter structs. All tools have descriptions (10-100 chars, mostly 40-80). Input schemas are well-defined with typed parameters. However, output schemas are NOT documented, responses are serialized to JSON text without declared structure. Error handling is minimal (generic McpError::internal_error). No tool annotations (readOnlyHint, destructiveHint). Parameter descriptions are terse (avg ~30 chars) and lack constraint details (ranges, enums, format hints). No pagination support despite tools like track_list and fx_list that could return large results. Composition is sound (single responsibility per tool), but missing recovery guidance in error messages.

• Add output schema documentation for all tools. Define response structure (e.g., track_list returns {tracks: [{index: int, name: string, volume_db: float, ...}], total: int}).
• Implement tool annotations in schema: add readOnlyHint=true for get_* and list_* tools; destructiveHint=true for remove_* and run_reascript; idempotentHint=true for stateless operations.
• Enhance parameter descriptions with explicit constraints: 'pan: float from -1.0 (left) to 1.0 (right)', 'pitch: integer 0-127 (MIDI note)', 'rgb: integer 0-255 per channel'.
• Add pagination to track_list and fx_list: accept optional limit (default 20, max 100) and offset parameters; return total count and next_offset in response.
• Improve error messages with recovery hints: 'Track index 5 out of bounds (0-3). Call track_list() to see available tracks.' instead of generic internal_error.