Safe Zhihu MCP
Provides read-only access to Zhihu content, enabling search and retrieval of questions, answers, articles, comments, replies, public user profiles, and public activities via an isolated browser profile.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Safe Zhihu MCP Search Zhihu for highly upvoted answers about learning Python"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Safe Zhihu MCP
safe-zhihu-mcp 是 alizeeblack-code/zhihu-mcp
的安全只读派生版。它通过独立的 Edge/Chrome 浏览器资料访问知乎,不需要知乎开放平台
Access Secret,也不会读取或导出日常浏览器 Cookie。
能力
搜索问题、回答和文章,支持排序与类型筛选
读取问题信息和可见高赞回答
读取回答全文与专栏文章全文
读取一级评论和楼中楼回复
读取用户公开资料与近期公开动态
用户明确要求时,将动态保存到本地 SQLite 供后续查询
Related MCP server: zhihu-mcp
硬性安全边界
MCP 中不存在发布、编辑、评论、回复、点赞、关注、上传或删除工具。
登录只能由用户在独立可见浏览器中亲自完成。
不读取 Edge/Chrome 日常资料,不导出
cookies.json。浏览器沙箱与 HTTPS 验证保持启用,拒绝跨域重定向。
主页面只能导航到
zhihu.com和zhuanlan.zhihu.com。所有 ID、用户名、关键词和 URL 均经过严格验证。
单并发访问,最短间隔 3 秒,每小时最多 60 次、每天最多 200 次。
输出限制为 500,000 字符,浏览器下载被禁用。
运行数据保存在操作系统用户数据目录,不进入仓库。
这些措施不能消除知乎验证码、限流或账号限制风险。建议使用副号并控制使用频率。
安装
uv tool install .
uv tool run --from playwright playwright install chromium
safe-zhihu-login第三步会打开独立浏览器。请亲自登录知乎,成功后回到终端按 Enter。
MCP 启动
safe-zhihu-mcp可用工具:
search_contentget_question_detailget_answer_detailget_article_detailget_commentsget_comment_repliesuser_profileget_activitiesget_new_sincecheck_login_status
get_activities(..., persist=false) 默认不保存。只有显式传入 persist=true 才写入本地数据库。
开发验证
uv sync --extra dev
uv run pytest -q许可证与来源
本项目保留原项目 MIT License 和作者版权。安全改造不代表知乎官方认可,也不改变知乎 网站与内容的使用规则。
Available Tools
10 toolscheck_login_statusB
Check the isolated browser profile without exposing cookies.
Inputs: no named parameters. Returns: result
get_activitiesC
Read recent public activity. Persistence is off unless explicitly requested.
Inputs: user (required), count, persist. Returns: result
get_answer_detailC
Read a full Zhihu answer.
Inputs: answer_id (required), question_id (required). Returns: result
get_article_detailC
Read a Zhihu column article.
Inputs: article_id (required). Returns: result
get_comment_repliesC
Read replies under one Zhihu comment.
Inputs: comment_id (required), count. Returns: result
get_commentsB
Read top-level comments for a question, answer, or article.
Inputs: url (required), count. Returns: result
get_new_sinceC
Query only locally persisted activity; does not access Zhihu.
Inputs: user, hours. Returns: result
get_question_detailC
Read question metadata and visible top answers.
Inputs: question_id (required). Returns: result
search_contentC
Search Zhihu questions, answers, and articles; read-only.
Inputs: keyword (required), sort, count, content_type. Returns: result
user_profileC
Read public profile information for one Zhihu user.
Inputs: token (required). Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 10 tools
Each tool targets a distinct resource (article, question, answer, comment, user, activity), so misselection is unlikely for the core reads. The only mild overlap is between get_activities (live read from Zhihu) and get_new_since (local cache only), which descriptions do clarify but an agent could still conflate.
Names are uniformly snake_case and mostly follow a get_<resource>_detail pattern (get_article_detail, get_question_detail, get_answer_detail, get_comments, get_comment_replies, get_activities). Deviations are minor: user_profile is a bare noun and search_content/check_login_status use different verbs, but all remain readable and predictable.
Ten tools is well within the ideal range and each maps to a concrete read operation on the Zhihu domain. Nothing feels redundant or padded for a read-only browsing server.
The read-only surface covers the main content lifecycle: search, question, answer, article, comments, replies, user profile, and activity, plus login status and incremental local activity. Minor gaps remain around enumerating a user's own answers/articles/followers or topic/collection listings, which agents would have to work around.
Maintenance
Related MCP Connectors
Zhihu/知乎 hot list, search/details, comments/replies, creators/articles, and video transcripts.
China news (EN/ZH), Weibo 微博/Baidu 百度/Zhihu 知乎 hot lists and deletion watch from public data.
China news (EN/ZH), Weibo 微博/Baidu 百度/Zhihu 知乎 hot lists and deletion watch from public data.
RedNote (Xiaohongshu, 小红书) note search, full notes, comment threads and creator profiles.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables AI assistants to search real content from Xiaohongshu and Zhihu, supporting keyword search, note details, and question retrieval.84MIT
- FlicenseNot gradedqualityDmaintenanceEnables AI assistants to directly operate Zhihu, including login, publishing articles and videos, searching content, getting recommendations, and commenting.7-
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to search, read, and analyze Zhihu content including questions, answers, comments, and user activities through the MCP protocol.5MIT
- FlicenseAqualityBmaintenanceRead-focused MCP server for Zhihu, enabling retrieval of pins, articles, comments, and replies, with optional limited write tools for drafting and editing answers.101-