SocialDataX 知乎 Zhihu MCP

Descriptions do a strong job of routing: URL-based extraction, keyword search, user lookup, and video speech-to-text each map to a distinct tool with explicit 'use X when you have Y' guidance. The main residual overlaps are the two near-parallel submit_video_speech_text tools (URL vs zvideo_id) and the subtle split between get_content_comments_by_url and get_comment_replies_by_url, which an agent could still confuse without careful reading.

The zhihu_* block is highly consistent: resource + verb + qualifier in snake_case (get_content_detail_by_url, search_users, submit_video_speech_text_by_video_url). The single socialdatax_get_points_balance follows the same snake_case verb_noun shape but uses a different service namespace, a minor deviation in an otherwise predictable scheme.

Thirteen tools is a well-scoped surface for a Zhihu data-extraction server: one detail reader, two comment readers, three search tools, two user tools, hot list, and a coherent three-tool async video pipeline. Each tool covers a distinct input mode or workflow step rather than duplicating functionality.

Core read workflows are covered end to end: content detail, comments plus replies with pagination, user profiles, search (content/users/suggestions), hot list, and video transcription with submit-and-poll. The notable gap is that user content lists stop at articles — no tool surfaces a user's answers, ideas, or videos, so retrieving a creator's non-article output requires a workaround.