Loading ToolBench
Read a private 404K Knowledge Planet archive through the Model Context Protocol
Server has well-structured tool definitions with complete input schemas using Zod validation and clear descriptions in Chinese. All 4 tools are READ_ONLY and properly guarded with error handling. However, descriptions lack English translations (limiting LLM accessibility), output schemas are not documented, and parameter descriptions could be more explicit about constraints and use cases. Tool naming follows verb_noun pattern (zsxq_status, zsxq_list_topics, zsxq_search_topics, zsxq_get_topic) which is clear. Schemas are comprehensive with enums, defaults, and validation rules, but lack documentation of return types.
按话题 ID 获取完整正文、分段作者、图片、PDF/附件本站地址和原文链接。
分页浏览已同步话题，支持标签、时间、精华、图片、附件、排除条件和排序。默认返回摘要；需要正文时使用 zsxq_get_topic。
搜索话题标题、正文和标签，并按精确标签、时间、精华、图片、附件、排除条件和热度组合筛选。可再按 ID 获取完整正文。
查看知识星球归档的最后同步时间、话题数量、媒体统计和源登录态剩余时间。
Output schemas not documented. LLMs cannot plan downstream calls or extract required fields without knowing what zsxq_status, zsxq_list_topics, and zsxq_get_topic return. Baseline: 100% of A+ tools document return types.
Descriptions are in Chinese only. LLMs trained primarily on English may misinterpret intent or fail to select the correct tool. Descriptions should be in English or bilingual to maximize LLM accessibility.
Parameter descriptions lack explicit constraint documentation. E.g., 'offset' and 'limit' have descriptions but no mention of min/max bounds in the description text itself (bounds are in Zod schema, not visible to LLM). Baseline: 94% of A+ tools have explicit constraint descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | B | 70 | 2026-07-28+ | v2 |
No documented pagination guidance in tool descriptions. zsxq_list_topics and zsxq_search_topics accept offset/limit but descriptions do not explain pagination behavior, total count availability, or when to stop iterating.
Error handling returns raw error messages without recovery guidance. toolError() wraps exceptions as plain text. LLMs need actionable next steps: 'Try search_topics() with a different query' or 'Check ZSXQ_API_TOKEN is set'.