What it does
It adds search tools to AI assistants such as Claude Code, Cursor and Codex, reaching sources that general web search does not cover. These include Bilibili videos and comments, Zhihu answers on request, recent V2EX threads and research papers, with links back to the original posts. Questions can be asked in English and answered from Chinese sources.
It suits research that depends on spoken or community content, such as finding what a speaker said in a video and at which minute. It can also track a topic through saved searches that report only new items, or map which papers a research field builds on and which recent work cites them. Audio from podcasts and videos can be transcribed locally, and results are kept in a local index.
It runs on the user's own computer and connects over MCP, so the AI app starts it when needed. It needs Python 3.11 or newer. It has no AI model of its own, no account and no telemetry, and the connected agent does the reasoning. Sites that require a login are off until the user enables them, and then use the login already in the user's browser.
Limitations and requirements
Requires Python 3.11 or newer. Podcast transcription, PDF reading and login-based sites are optional add-ons. Sites that need a login stay off until enabled and then use the login already in the user's own browser.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
OmniSeek
Give your AI agent what web search doesn't reach: podcasts and videos, Chinese communities, papers.
v0.3.0As shown on registry.modelcontextprotocol.io when we last read it