Server has 10 tools with basic schemas and descriptions, but significant quality gaps. Most tool descriptions are under 50 chars (well below the 194-char baseline for A+ tools). Parameter descriptions are minimal or absent. No output schemas documented. Error handling is generic ('Device not found') with no recovery guidance. The 'validate' tool has an empty schema and cryptic purpose. TV-specific tools (tv_volume_up, tv_volume_down, tv_mute, tv_open_app) lack parameter constraints (e.g., app enum for tv_open_app). No tool declares idempotency, destructiveness, or read-only hints. Room/device name resolution is undocumented in tool descriptions, forcing LLMs to guess. Overall, definitions are functional but far below production quality.

• Expand all tool descriptions to 100 - 200 chars. Include WHAT (action), WHEN (use case), and WHAT IT RETURNS. Example: 'Turn on a device in a room. Idempotent, safe to retry. Returns success message or error if device not found. Use get_all_states() first to discover available devices and rooms.'
• Add enum constraint to tv_open_app 'app' parameter: {"type": "string", "enum": ["netflix", "youtube", "home"], "description": "App to open on TV (netflix, youtube, or home)"}
• Document output schemas in tool descriptions. Example for get_device_state: 'Returns JSON object with fields: status (on/off), brightness (0-100 for lights), temperature (for AC), etc. See get_all_states() for full device structure.'
• Add error recovery guidance. When device not found, return: 'Device "living_room/tv" not found. Available devices in living_room: [tv, light, fan]. Call get_all_states() to see all devices.'
• Add tool annotations (destructiveHint, readOnlyHint, idempotentHint) to help LLMs reason about side effects. Example: turn_on_device should have destructiveHint=false (idempotent), get_device_state should have readOnlyHint=true.