0xmariowu/Autosearch · github ↗ · pipeline

◑ WATCH category agent-framework maturity active license MIT apply project seen 2026-08-17 → 2026-08-17 #mcp#research#search#channels#agents#chinese-sources

0xmariowu/Autosearch

MCP-native search API across 40 channels including 10+ Chinese social platforms.

What it does

40-channel search access — unified queries across arxiv, GitHub, Reddit, Hacker News, WeChat, Twitter/X, Bilibili, Xiaohongshu, Weibo, Douyin, Zhihu, and 30+ others
MCP-native integration — plugs into Claude Code, Cursor, Cline, GPT-Researcher, LangChain, LlamaIndex as an MCP server
No LLM coupling — returns indexed, deduplicated, ranked results; your agent keeps its own model and prompts
Chinese social media first — 10+ Chinese sources (WeChat, Bilibili, Weibo, Douyin, Zhihu, Xiaohongshu, Xueqiu) covering product reviews, tech discussion, financial forums
Zero-config for 27 channels — arxiv, GitHub, DuckDuckGo, Hacker News, Reddit, Stack Overflow ready to use; YouTube, TikHub-gated Chinese platforms, Xueqiu unlock with API keys or CLI login

How it works

Node.js CLI that exposes 40 channels as an MCP server. Agents query via MCP; Autosearch fans requests across channels (academic databases, GitHub, social platforms, forums) and returns ranked, deduplicated results with source URLs. Channel-specific adapters handle APIs: TIKHUB for TikTok/Bilibili/Weibo, Jina for LinkedIn, native parsers for public forums. Stack: Node.js, MCP protocol, channel-specific SDK wrappers; some Chinese channels require paid API keys.

Standing & signals

Recently active (pushed 2026-08-07), published on npm with passing CI. 42 stars indicates early adoption, not yet mainstream. Comprehensive documentation (channel matrix, MCP setup, examples) with Chinese README signals Asia-market ambition. Positioning as "deep research for AI agents" filling gaps where LLMs have no API access (Twitter/X, Reddit, GitHub, Chinese social) is credible and addresses real agent pain.

Q2 — Apply where

Project (Research Console). The console crawls trending sources and categorizes them. Autosearch could supplement the fetch layer with Twitter/X (via TIKHUB) and Chinese social media (Weibo, Douyin, Zhihu, WeChat) without building channel adapters per-source.

Q3 — Fit for current projects + sketch

The Research Console generates daily reports from crawled sources. Autosearch is a search API, not a reporting pipeline — but could feed the categorization layer. Integration: (1) wire Autosearch as an MCP server in the console's constitution, (2) add a parallel fetch stage querying Autosearch for trending topics, (3) deduplicate across the console's existing crawlers. This unlocks 10+ Chinese social channels and Twitter/X without individual integrations. Caveat: TIKHUB (TikTok, Bilibili, Weibo, Douyin) is a paid tier; marginal value for research-trending (where news-focused sources already dominate) unless ClientCo pivots to real-time market-sentiment monitoring.

Q4 — Future-watch verdict

WATCH. Active development, solid channel breadth (especially Chinese platforms), MCP-native design, and documented setup make it a credible research-data source for future upgrades. Not adopted now (console crawlers work), but worth monitoring if real-time social sentiment becomes a ranking signal or if ClientCo starts tracking Chinese product-launch discussion.