# lilbee

> lilbee is an open-source MCP server for files and documents. Local AI search over your files, code and crawled sites, with cited answers and a built. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** Web, macOS, Windows, Linux, CLI, API, self-hosted, vscode
- **Website:** https://lilbee.sh
- **Repository:** https://github.com/tobocop2/lilbee
- **MCP Registry name:** io.github.tobocop2/lilbee
- **Categories:** files and documents
- **Listing updated:** 2026-11-18

## About

Rather than a model server you wire up yourself, lilbee is a single executable that downloads models from Hugging Face, runs them on local GPUs (Metal, Vulkan, CUDA or ROCm), and places larger models across several cards. Its search indexes folders of PDFs, notes, ebooks, code and scanned images, and every answer links back to the file and line or page it came from.

The README describes a few extras that go beyond plain document chat: a crawler that saves whole docs sites as markdown for offline search, an MCP server that lets coding agents query the library, a launcher that starts opencode, Claude Code, pi or Hermes on local models, and an optional memory feature that is off by default. Existing Ollama and LM Studio installs can be used as backends.

It suits people who want document search and chat on their own hardware without Docker or a separate database, and developers who want their coding agent to answer from a local codebase and docs. The project describes itself as beta, with pre-releases on PyPI and no code signing on macOS arm64 and Windows builds.

## Features

- **Built-in model manager:** Browses Hugging Face, downloads GGUF models, assigns them roles and runs them on the local GPU.
- **Cited answers:** Responses name the source file and line or PDF page, which can be clicked to open the source.
- **Multi-GPU placement:** Splits a large chat model across several cards and copies the embedder to each one.
- **Website crawler:** Fetches pages, including recursive crawls, into markdown in the library for offline search.
- **MCP server for agents:** Exposes search over the Model Context Protocol so MCP-aware coding agents can cite your files.
- **Code-aware chunking:** Splits code by function and class using tree-sitter rather than fixed line ranges.

## Good for

- Search a folder of manuals, PDFs or research papers and get answers with page citations
- Give a coding agent a local reference of project code, dependencies and API docs
- Keep an offline copy of a documentation site or wiki that is searchable without internet
- Run a local chat model on an existing GPU without setting up Docker or a model server
- Index scanned documents through OCR or a local vision model for search

## Install

### Claude Code (Terminal)

```sh
claude mcp add lilbee -- npx -y lilbee
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "lilbee": {
      "command": "npx",
      "args": [
        "-y",
        "lilbee"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "lilbee": {
      "command": "npx",
      "args": [
        "-y",
        "lilbee"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "lilbee": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "lilbee"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "lilbee": {
      "command": "npx",
      "args": [
        "-y",
        "lilbee"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "lilbee": {
      "command": "npx",
      "args": [
        "-y",
        "lilbee"
      ]
    }
  }
}
```

## FAQ

### Does my data leave my machine?

The README says files, the index and questions stay local, and a cloud model is used only when one is selected.

### Which platforms are supported?

Prebuilt bundles exist for macOS, Windows and Linux, with Metal, Vulkan, CUDA and ROCm variants, plus npm, PyPI and Docker options.

### Is lilbee open source?

The repository is published under the MIT licence.

### Is lilbee free?

Yes. lilbee is open source, so you can use and self-host it for free.

### How do I add lilbee to Claude Code?

Run this command in your terminal: claude mcp add lilbee -- npx -y lilbee

### What are the best alternatives to lilbee?

The closest alternatives to lilbee are SSRS2, ASTra, CARTESIAN ECU, Invoices and SigPen.

## Where this comes from

- Description: written from the project's README, the project's website and its MCP Registry entry.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [SSRS2](https://ai-powered.com/mcp/ssrs2): SSRS2 is an MCP server for files and documents.
- [ASTra](https://ai-powered.com/mcp/astra): ASTra is a freemium MCP server for files and documents.
- [CARTESIAN ECU](https://ai-powered.com/mcp/cartesian-ecu): CARTESIAN ECU is a paid MCP server for files and documents.
- [Invoices](https://ai-powered.com/mcp/invoices): Invoices is a paid MCP server for files and documents.
- [SigPen](https://ai-powered.com/mcp/sigpen): SigPen is a freemium MCP server for files and documents.
- [ConvertAudioToText](https://ai-powered.com/mcp/convertaudiototext): ConvertAudioToText is a freemium MCP server for design and media.
- [WhoScribe](https://ai-powered.com/mcp/whoscribe): WhoScribe is a freemium MCP server for design and media.
- [filex](https://ai-powered.com/mcp/filex): filex is an open-source MCP server for files and documents.

---
Source: https://ai-powered.com/mcp/lilbee
