Loading ToolBench
用於取得臺灣中央氣象署 API 資料的 Model Context Protocol (MCP) Server
Server has one tool with complete input schema and reasonable description. However, the description, while localized to Taiwanese Chinese, lacks strategic context about when/why to use this tool, and parameter descriptions are minimal. The tool name follows verb-noun convention appropriately. Output schema is implicitly structured (JSON response from weather API) but not explicitly documented in the tool definition. Error handling exists but is generic. The server is straightforward but lacks the depth expected of production-grade weather tools.
取得臺灣各縣市天氣預報資料 - 今明 36 小時天氣預報。 (可用縣市名稱： 宜蘭縣、花蓮縣、臺東縣、澎湖縣、金門縣、連江縣、臺北市、新北市、桃園市、臺中市、臺南市、高雄市、基隆市、新竹縣、新竹市、苗栗縣、彰化縣、南投縣、雲林縣、嘉義縣、嘉義市、屏東縣)
Tool description lacks context about output structure and use cases. Current description lists valid locations but does not explain what forecast data is returned (e.g., temperature, precipitation, wind speed), time horizons, or when to call this tool vs alternatives.
Parameter 'locationName' has no description in the schema. The enum constraint is present but the parameter description field is empty. LLMs cannot reason about why this parameter exists or what it controls without explicit description text.
Output schema is not documented. Tool definition does not specify what fields the weather forecast response contains (temperature range, precipitation probability, wind direction, etc.), forcing LLMs to infer structure from raw JSON.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 48 | - | v1 |
Error handling returns generic error messages without recovery guidance. Example: 'Invalid input: {...}' does not tell the LLM what to do next or which location names are valid (though enum is defined, the error should reference it).
STDIO-only transport limits accessibility. Server cannot be used with hosted MCP clients or remote agents, constraining production deployment scenarios.