Six tools with Chinese descriptions and basic parameter schemas. All tools have descriptions (10-200 chars, mostly 50-150), but lack English translations and depth. Input schemas present with type declarations and descriptions, but output schemas are undocumented (returns raw API responses). No error recovery guidance, no parameter constraints (enums, ranges), no tool annotations. Parameters accept free-form strings where enums would prevent hallucination (e.g., 'mode' in map_distance_matrix accepts 'driving|riding|walking' but not declared as enum). Tool names follow verb_noun pattern (map_*) but descriptions are domain-specific Chinese without LLM-optimized English. No pagination, no result limits documented. Composition is reasonable (6 focused tools), but chaining support unclear, no documented output fields that feed into downstream tool inputs.

• Document output schemas for all tools. Specify which fields are returned (e.g., map_geocode returns {status, result: {location: {lat, lng}, precise, comprehension, confidence}}). Use JSON Schema or a structured format.
• Convert parameter constraints to enums. Replace 'mode: string' with 'mode: enum[driving|riding|walking|transit]' in map_directions and map_distance_matrix. Add range constraints to numeric params (e.g., radius: int, min=1, max=50000).
• Provide English descriptions with LLM-optimized guidance. Expand each description to 100-200 chars. Include: WHAT (action), WHEN (use case), PREREQUISITES (e.g., 'Call map_geocode first if you only have an address'), and RETURNS (key fields).
• Add error recovery hints. When API returns status != 0, return actionable messages: 'Address not found. Try a more complete address (city, district, street) or use map_search_places to find nearby POIs.' Categorize errors as retryable vs user-fixable.
• Implement result limits and pagination. Add 'page_size' (default 20, max 50) and 'page_num' params to map_search_places. Return 'total_count' and 'has_more' in response. Document limits in tool descriptions.