Server has 6 tools with complete input schemas and descriptions in Traditional Chinese. Naming follows verb_noun pattern (search_by_channel, recommend_payment, compare_cards, get_promotions, get_card_details, list_all_cards). Descriptions are detailed and context-aware (avg ~180 chars), explaining WHEN to use each tool. However, output schemas are not documented in the visible code, tool_trace.py shows compact display logic but no formal schema definitions. Parameters are well-described with enums (channel IDs), defaults, and constraints. Error handling is minimal, no recovery guidance or error categorization visible. All tools are READ_ONLY (safe), but no explicit error responses documented.

• Document output schemas for all 6 tools in JSON Schema format. Include field names, types, and descriptions. Example: search_by_channel returns {results: [{rank, card_id, card_name, cashback_rate, estimated_cashback, conditions}], channel_id, channel_name, error?}.
• Add explicit error handling: return structured errors with error_code, message, and recovery_hint. E.g., {error: 'INVALID_CARD_ID', message: 'Card ctbc_xyz not found', hint: 'Call list_all_cards() to see available cards'}.
• Add a 'list_channels' tool (no params) that returns all 17 valid channel IDs with human-readable names. This enables discovery and prevents LLM from guessing invalid channels.
• Validate 'cards_owned' array at tool entry and return actionable error: 'cards_owned cannot be empty. Call list_all_cards() first to see available cards.' instead of silent failure.
• Provide English translations of all tool descriptions and parameter docs. Use a bilingual schema or add an 'en' field alongside 'zh' descriptions for developer clarity.