Loading ToolBench
MCP server for Zhihu (知乎) - AI-driven content automation via browser
Server has 11 tools with complete input schemas and descriptions in Chinese. However, descriptions lack LLM-optimization guidance (no WHEN to use, no prerequisites, no recovery hints). Parameter descriptions are present but minimal (avg ~40 chars vs baseline 72). No output schemas documented. Error handling is absent, tools return raw dicts without structured error classification or recovery guidance. Naming is verb-first and clear (check_, get_, delete_, publish_, search_, post_, reply_), but composition issues exist: publish_article and publish_video could be unified; post_comment and reply_comment are similar but distinct. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Security concern: no evidence of input validation or sanitization for user-provided content (title, content, tags, URLs). Descriptions are in Chinese, which may reduce LLM comprehension in English-first agents.
检查知乎登录状态。无参数，直接调用即可返回当前是否已登录及用户名。
删除 cookies 文件，重置登录状态。删除后需要重新扫码登录。
获取知乎帖子详情（包括互动数据和评论）。参数：url（帖子完整 URL，必填，从搜索结果或推荐列表获取）、load_comments（是否加载评论，默认 true）、comment_limit（最大评论数，默认 20）。
获取知乎登录二维码。返回 Base64 编码的二维码图片，用户需使用知乎 App 扫码登录。
获取知乎首页推荐列表。参数：scroll_count（滚动加载次数，默认 3）、limit（最大返回条数，默认 20）。无需登录也可使用，但登录后推荐更精准。
获取知乎用户个人主页信息。参数：user_url（用户主页完整 URL，必填，格式如 https://www.zhihu.com/people/xxx）。返回用户基本信息、粉丝数、关注数、最近内容等。
发表评论到知乎帖子。参数：url（帖子完整 URL，必填）、content（评论内容，必填）。需要先登录才能发表评论。
No output schemas documented. LLMs cannot predict response structure, forcing them to guess at field names for downstream tool chaining. E.g., publish_article returns dict but no schema shows what fields are present (success, message, article_id, url, etc.).
Descriptions lack LLM-optimization guidance. Missing WHEN to use, prerequisites (e.g., 'requires login'), and recovery hints. E.g., post_comment says 'needs login' but doesn't guide LLM to call check_login_status first or handle auth errors.
No error handling or recovery guidance. Tools return raw dicts without error classification (retryable vs fatal). E.g., if publish_article fails due to auth, network, or validation, LLM has no guidance on what to do next.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
发布图文内容到知乎。参数：title（标题，必填）、content（正文内容，必填）、images（图片路径列表，可选，支持本地绝对路径）、tags（话题标签列表，可选）。示例：title='我的文章', content='正文内容...', images=['/path/to/img.jpg']
发布视频内容到知乎。参数：title（标题，必填）、content（视频描述，必填）、video_path（本地视频文件绝对路径，必填）、tags（话题标签列表，可选）。仅支持本地视频文件，不支持 URL。
回复知乎帖子下的指定评论。参数：url（帖子完整 URL，必填）、comment_author（要回复的评论作者昵称，必填）、content（回复内容，必填）。需要先登录才能回复评论。
搜索知乎内容。参数：keyword（搜索关键词，必填）、search_type（搜索类型：综合/问题/回答/文章/用户，默认综合）、limit（最大结果数，默认 20）。
No input validation or sanitization visible. User-provided strings (title, content, tags, URLs) are passed directly to browser automation without validation. Risk of injection attacks or malformed input causing silent failures.
Descriptions in Chinese may reduce comprehension in English-first LLM agents. Tool names are English but descriptions are Chinese, creating a mismatch that could confuse multilingual models or English-only agents.