# AutoRAG [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/Marker-Inc-Korea/AutoRAG  
**GitHub Stars:** 6008  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/autorag

## Description
Your Personal Librarian Agent. Ask anything on your computer, organize files, remember important info and files.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "autorag": {
    "command": "npx",
    "args": ["-y","-p","@autorag/librarian","autorag-mcp"]
  }
}
```

## Documentation

## What AutoRAG does

AutoRAG provides an agent-facing way to search and work with information stored on a computer. Its primary use is finding relevant content inside local files and returning answers tied to the source passages. The wider AutoRAG system also supports file organization workflows, indexing selected folders, duplicate detection, and searches across connected data sources.

The repository distinguishes the newer AutoRAG Agent from the earlier Python AutoRAG project. The newer system is focused on personal knowledge and office-work retrieval; the legacy project remains available under `legacy/` for RAG pipeline optimization.

## How it works

Search combines traditional keyword matching with meaning-based retrieval and a reranker. This lets an agent find material by exact terms as well as by the subject or idea expressed in a document. Answers can include links to the passages that support them.

Semantic indexing runs locally through MinSync's Qwen3-Embedding-0.6B engine with 1024-dimensional embeddings. File contents are not sent to a remote embedding service unless a remote `minSync.embedder` is configured. Credentials and API keys are stored in the operating system's encrypted store rather than in a configuration file.

Indexed content can include local documents, synced folders from services such as Google Drive, iCloud Drive, OneDrive, and Dropbox, plus sources including Slack, KakaoTalk, Discord, Telegram, WhatsApp, mail, Notion, GitHub, Obsidian, browser history, RSS, photos, and coding-agent sessions. The available sources depend on the corresponding setup.

## Setup and configuration

The project supports agent-assisted setup through the `autorag-setup` and `autorag-lite-setup` skills. These workflows ask which folders to index, configure the relevant search setup, and run a health check and test search. The Lite setup is intended to provide model-free search over selected files through an MCP server.

The setup instructions explicitly caution against moving, renaming, editing, or deleting documents during installation. A separate `autorag-doctor` skill checks the installation and reports per-datasource status.

The repository lists Node.js 24 or newer and Bun 1.2 or newer as supported runtimes. Model access can use the hosted Dazzi Models plan or a provider key supplied by the user.

## Tools and capabilities

- Search indexed files using keyword and semantic matching.
- Retrieve answers grounded in cited source passages.
- Index chosen folders and search synced cloud-drive content.
- Search supported messaging, mail, notes, GitHub, browser, RSS, photo, and coding-session sources.
- Support file-triage and organization workflows in the broader AutoRAG application.
- Re-index on a schedule, including while the desktop application is closed.

The repository also includes Dazzi, a desktop application built around the same engine. Dazzi adds file browsing, previews, duplicate grouping, and remote access; these are application features rather than requirements for using the agent-facing search setup.

_Full upstream README: https://allmcps.com/mcp/autorag/readme_

