# AutoSearch

Open-source deep-research infrastructure for AI agents that searches 40 channels, including more than ten Chinese sources, and returns indexed, deduplicated multi-source results while staying decoupled from any specific LLM provider.

AutoSearch plugs into agent hosts a developer already uses, including Claude Code, Cursor, Cline, GPT-Researcher, LangChain, LlamaIndex and AutoGen, adding MCP-native access across 40 channels such as arXiv, PubMed, OpenAlex, DBLP and GitHub with no configuration needed, plus channels that unlock after additional setup. Because the engine only returns indexed, ranked and deduplicated results with a source URL for each, rather than making its own model calls, the calling agent keeps using its own model, prompts and workflow.

It installs through an npx command, a script pasted into an AI agent for guided setup, or a shell install script, and deliberately avoids npm install-time scripts as a supply-chain hardening measure, so a plain npm install does not auto-run its own setup step. Once installed, an agent can search all 40 channels at once, including academic papers, developer communities, Chinese social media platforms and forum threads that would otherwise require separate manual searches across each platform.

It is written in Python and released under the MIT license, with documentation offered in English and Chinese.

Website: https://www.autosearch.dev
