Loading ToolBench
Single-tool server with decent fundamentals but notable gaps. The get-forecast tool has a clear verb-noun name, proper parameter schema with min/max bounds, and a reasonable description. However, the server lacks structured output schema documentation, error handling guidance, and the description is in Chinese (barrier for English-speaking LLMs). Output formatting is hardcoded as markdown text rather than structured JSON. No tool annotations (readonly hints, idempotency indicators). Parameter descriptions are adequate but sparse. Overall falls into the 'Fair' band, solid basics, but missing polish and LLM-optimized guidance patterns.
获取天气预报
Tool description is in Chinese ('获取天气预报') rather than English. LLMs trained primarily on English will not understand when/why to select this tool. Non-English descriptions severely limit discoverability and reasoning in English-primary agents.
No documented output schema. The tool returns hardcoded markdown text via a single 'text' content block. LLMs cannot reason about downstream data extraction or chaining because they don't know the structure of the response. Should document: what fields are in the daily forecast, what timezone dates use, what units temperatures/wind use.
No tool annotations. The tool has no readOnlyHint (it is read-only), no idempotentHint (calling twice with same coords returns same data), and no destructiveHint (N/A). These hints enable agents to reason about safe retry and caching behavior.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 41 | - | v1 |
Error handling lacks recovery guidance. When API fails (response === null), error message says 'Failed to retrieve coordinates: {lon}, {lat}. Response: null' but does not suggest what the agent should do next: retry? check the API key? use different coordinates? The error is not actionable.
Parameter descriptions are minimal (11-21 chars each). 'The longitude of the location' and 'The latitude of the location' meet the bare minimum but don't explain expected precision, when to use them, or dependencies (both required together). Baseline for A+ tools is ~72 chars per parameter.