Two read-only tools with minimal schemas and generic descriptions. Tool naming follows verb_noun convention but descriptions lack depth needed for LLM selection. Input schemas present but sparse, get_usage accepts one optional parameter with minimal constraint info; get_kimi_usage has empty input schema. No output schemas documented. Error handling guidance absent. No parameter descriptions beyond a single optional provider field. Descriptions are bilingual but generic, not optimized for LLM reasoning about when to invoke each tool.

• Add explicit enum constraint to get_usage 'provider' parameter: {"enum": ["kimi", "openai", "anthropic", "openrouter"]} with description 'Filter results to a specific LLM provider, or comma-separated list of providers. Omit to fetch all configured providers.'
• Expand tool descriptions to 50 - 150 characters, explaining WHEN to use each: e.g., 'Get current Kimi Coding Plan API quota and usage. Call this for real-time remaining requests/tokens on Kimi. Use get_usage() to compare quotas across all configured providers.' and 'Fetch API usage statistics for one or more LLM providers (Kimi, OpenAI, Anthropic, OpenRouter). Returns quota limits, used amount, and remaining requests/tokens per provider.'
• Document output schemas in tool descriptions or as structured examples: e.g., for get_usage() return {"provider": string, "quota_tokens": number, "used_tokens": number, "remaining_tokens": number, "reset_at": ISO8601_date}
• Add error recovery guidance: 'If authentication fails (401): check KIMI_API_KEY env var is set. If provider service is unavailable (503): retry in 60 seconds. If the response is empty: verify the provider is configured in the server settings.'
• Add readOnlyHint annotations to both tools in MCP tool definition (if using MCP 2026-07-28 format with tool.annotations).
• Distinguish use cases in descriptions: 'get_usage() is for discovering what quota limits are available across providers; get_kimi_usage() is for rapid polling of Kimi-only real-time quota. If you only care about Kimi, prefer get_kimi_usage() for lower latency.'