> ## Documentation Index
> Fetch the complete documentation index at: https://docs.autosearch.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# AutoSearch

> Deep research MCP server for AI developers — 40 channels, cited results, no hallucination

AutoSearch is an MCP server that gives your AI agent access to 40 real search channels: academic databases, developer platforms, Chinese social media, and Linux DO forum coverage. Every result is cited and deduplicated. No hallucination.

Works with **Claude Code**, **Cursor**, and any MCP-compatible client.

## Install

<CodeGroup>
  ```bash npx (recommended) theme={null}
  npx autosearch-ai
  ```

  ```bash curl theme={null}
  curl -fsSL https://raw.githubusercontent.com/0xmariowu/Autosearch/main/scripts/install.sh | bash
  ```

  ```bash pip theme={null}
  pip install autosearch
  autosearch init
  ```
</CodeGroup>

`init` auto-detects your LLM providers and registers AutoSearch with your MCP-capable AI clients automatically. For Claude Code that uses `claude mcp add` (writing to `~/.claude.json`); for Cursor / Zed it writes to the per-client MCP config file.

## What you get

<CardGroup cols={2}>
  <Card title="40 Search Channels" icon="magnifying-glass">
    arxiv, PubMed, GitHub, Hacker News, Reddit, Linux DO, Xiaohongshu, Weibo, Twitter, Douyin, and more — English and Chinese, free and paid tiers.
  </Card>

  <Card title="22 MCP Tools" icon="wrench">
    `run_channel`, `delegate_subtask`, `consolidate_research`, citation management, deep research loops, and more.
  </Card>

  <Card title="5 Search Modes" icon="sliders">
    academic / news / chinese\_ugc / developer / product — auto-detected from the query, or set manually.
  </Card>

  <Card title="Cited Results" icon="link">
    Every result includes source URL. No invented facts. Runs BM25 + SimHash dedup to surface the best results.
  </Card>
</CardGroup>

## Quick example

Once AutoSearch is running as your MCP server, your agent can do:

```
run_clarify("latest transformer architectures for long-context 2026")
→ delegate_subtask(["arxiv", "papers", "hackernews"], query)
→ consolidate_research(evidence)
→ citation_export(format="markdown")
```

## Next steps

<CardGroup cols={2}>
  <Card title="Installation" icon="download" href="/install">
    Full install guide — MCP config, optional channels, troubleshooting
  </Card>

  <Card title="Channels" icon="list" href="/channels">
    All 40 channels with tier, languages, and requirements
  </Card>

  <Card title="MCP Clients" icon="plug" href="/mcp-clients">
    Config samples for Claude Code, Cursor, and others
  </Card>

  <Card title="Quickstart" icon="rocket" href="/quickstart">
    First search in 5 minutes
  </Card>
</CardGroup>