Server has 7 tools with basic descriptions and Zod schemas. However, most descriptions are under 50 chars (well below the 194-char baseline), lack context on WHEN to use each tool, and omit prerequisites or state-change implications. Parameter descriptions are minimal or absent. Tools like 'discipline', 'give_medicine', 'clean_poop', and 'toggle_lights' have empty input schemas with no parameters documented. The 'feed' and 'play_game' tools have Zod schemas but lack parameter descriptions in the code. Error handling returns JSON but provides no recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite 6 of 7 being WRITE operations. Output schema is undocumented, callers cannot predict response structure.

• Add parameter descriptions to 'feed' and 'play_game' schemas. E.g., for feed: 'type: The food category, meal provides full nutrition, snack provides partial nutrition and happiness boost.'
• Expand all tool descriptions to 100 - 200 chars. Include WHEN to use, prerequisites, and side effects. E.g., 'Discipline the pet when a discipline call is pending (check get_status for pending_discipline flag). Reduces happiness but improves obedience. Only works once per pending event.'
• Add tool annotations: mark feed, play_game, discipline, give_medicine, clean_poop, toggle_lights with destructiveHint=true (they modify pet state); mark get_status with readOnlyHint=true.
• Document the output schema in tool descriptions or a separate schema file. E.g., 'Returns {state: {hunger: 0-100, happiness: 0-100, health: 0-100, age: number, ...}}' so LLMs know what fields to expect.
• Enhance error messages with recovery guidance. E.g., 'Discipline failed: no pending discipline call. Call get_status to check pet state and available actions.' instead of bare error text.