Server has 6 tools with basic schemas and descriptions in Chinese. All tools have input schemas with type definitions and parameter descriptions. However, descriptions are inconsistent in depth and clarity for LLM reasoning. Output schemas are not formally documented, responses are inferred from code. Tool naming follows verb_noun pattern (browse_, list_, plant_, explore_, search_, live_) which is good. Critical gap: no error handling guidance, no recovery instructions, and no output schema documentation. The plant_idea tool accepts secrets implicitly (auto-registration) but does not document this flow clearly. browse_garden and explore_graph have vague descriptions that don't explain when to use each vs. the other.

• Document output schemas for all tools. For browse_garden, declare: returns list of {id: int, title: str, category: str, cross: bool, children: int, depth: int}. For plant_idea, declare: {status: str, produced_title: str, produced_text: str, latency_ms: int, tip: str}. This enables LLMs to plan chained calls.
• Add error handling and recovery guidance. Wrap r.raise_for_status() in try-except blocks and return structured errors like {error: 'Service unavailable. Retry in 30s.'} or {error: 'Node not found. Try search_ideas() to find similar nodes.'}
• Clarify browse_garden vs explore_graph in descriptions. E.g., 'browse_garden: Quick discovery of trending/recent nodes. explore_graph: Trace lineage and source layers for a specific node or all sources.'
• Document the auto-registration side effect in plant_idea description: 'First call auto-creates an anonymous account if AGENTPAY_API_KEY is not set. Subsequent calls reuse the same account.'
• Provide English descriptions alongside Chinese, or translate to English. If Chinese is required, use formal enum declarations (e.g., sort: enum[hot, latest, cross]) instead of inline examples.