Server has 4 tools with complete input schemas and descriptions in Japanese. Naming follows verb_noun pattern (create_plan_with_steps, start_step, complete_step, get_progress). Descriptions are detailed (200-400 chars) and include usage guidance. However, output schemas are not formally documented in code, responses are inferred from implementation. Parameter descriptions are present but lack explicit constraints (enums, ranges). Error handling is minimal; no recovery guidance or error categorization visible. Tool composition is sound (each does one thing), but missing idempotency guarantees and rate-limit documentation.

• Add explicit output schema documentation for each tool. Use TypedDict or docstring format to declare return fields, types, and meanings. Example: 'Returns: {plan_id: str, topic: str, total_steps: int, steps_summary: List[{order: int, title: str}], next_action: str}'
• Implement input validation with clear error messages. Example: 'step_order must be 1-based integer between 1 and {total_steps}. Got: {step_order}'. Return error responses that guide LLM recovery.
• Add parameter constraints to schema. Use minLength/maxLength for strings, minimum/maximum for integers, minItems/maxItems for arrays. Example: steps array minItems=5, maxItems=10.
• Document idempotency behavior. If create_plan_with_steps should be idempotent (same topic+steps = same plan_id), implement deduplication or state this clearly. If not idempotent, warn LLMs to avoid retries.
• Translate tool descriptions and server instructions to English while preserving Japanese as optional. This enables international collaboration and maintenance.