A Tamagotchi-like pet MCP server for LLM interaction research
Server has 7 tools with minimal schema and description quality. Six simple pet-care tools (observe_pet, feed_pet, play_with_pet, heal_pet, put_to_sleep, give_treat) have empty input schemas {} and generic 1-2 sentence descriptions. The run_pet_game tool has a proper schema with 4 parameters and descriptions, but overall the server lacks LLM-optimized descriptions (target 50-200 chars), parameter-level documentation, output schema definitions, and error handling guidance. No tool annotations (readOnlyHint/destructiveHint). Descriptions are too brief to guide LLM tool selection effectively.
Feed the pet: decreases hunger by 10, increases happiness by 5.
Give treat: small happiness boost (+5).
Heal the pet: decreases sickness by 15.
Get current pet state (hunger, happiness, sickness, energy, turn).
Play with pet: increases happiness by 10, decreases energy by 10.
Put pet to sleep: increases energy by 20.
Run a full ReAct loop session for pet care. An LLM will interact with the pet tools in a ReAct loop, observing the pet's state and taking actions to keep it healthy.
Six tools have empty input schemas ({}).
Descriptions for 6 simple tools are 1-2 sentences (10-30 chars), below the 50-200 char baseline for LLM-optimized descriptions.
No output schemas documented. LLMs cannot infer what fields to expect from tool responses (e.g., observe_pet returns JSON but structure is not declared).
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 37 | 2026-07-28+ | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). observe_pet is read-only; feed_pet, play_with_pet, heal_pet, put_to_sleep, give_treat are destructive writes. Annotations help LLMs reason about side effects.
No error handling or recovery guidance. Tools return JSON state on success but no documented error cases, validation rules, or actionable error messages for LLM recovery.