# haiku.rag — Agentic RAG CLI + MCP Server

> haiku.rag is an agentic RAG toolkit with CLI, Python API, and MCP server; verified 524★ and supports `add-src`, `ask --cite`, and `serve --mcp`.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

```bash
pip install haiku.rag
haiku-rag add-src paper.pdf
haiku-rag search "attention mechanism"
haiku-rag ask "What datasets were used for evaluation?" --cite
haiku-rag serve --mcp --stdio
```

## Intro

haiku.rag is an agentic RAG toolkit with CLI, Python API, and MCP server; verified 524★ and supports `add-src`, `ask --cite`, and `serve --mcp`.

**Best for:** Teams building citation-heavy RAG with local-first LanceDB storage and agent workflows

**Works with:** Python 3.12+ plus an embedding provider (Ollama/OpenAI/etc.) as required by README

**Setup time:** 6-15 minutes

### Key facts (verified)

- GitHub: 524 stars · 35 forks · pushed 2026-05-13.
- License: MIT · owner avatar + repo URL verified via GitHub API.
- README-backed entrypoint: `haiku-rag serve --mcp --stdio`.

## Main

- Start with one PDF and verify citations (`--cite`) before scaling to directory monitoring or research agents.

- Use the MCP server mode when you want assistants like Claude Desktop to manage documents/search/QA as tools rather than pasted context.

- Keep provider swaps explicit: embeddings and QA models are pluggable; document which provider you used for each dataset to make runs reproducible.

### Source-backed notes

- README states it is built on LanceDB, Pydantic AI, and Docling, and includes both CLI and Python API entrypoints.
- README documents MCP server usage: `haiku-rag serve --mcp --stdio` and a sample `mcpServers` JSON config.
- README lists multiple features including hybrid search, citations with page numbers/section headings, and local-first embedded LanceDB storage.

### FAQ

- **Do I need an embedding provider?**: Yes — README says you must configure one (Ollama/OpenAI/etc.) before indexing/searching.
- **Can I use it from an MCP client?**: Yes — run `serve --mcp --stdio` and add it to your client config.
- **Is there a slim install?**: Yes — README mentions `haiku.rag-slim` plus extras; use it when you want fewer deps.

## Source & Thanks

> Source: https://github.com/ggozad/haiku.rag
> License: MIT
> GitHub stars: 524 · forks: 35

---

<!-- ZH -->

## Quick Use

```bash
pip install haiku.rag
haiku-rag add-src paper.pdf
haiku-rag search "attention mechanism"
haiku-rag ask "What datasets were used for evaluation?" --cite
haiku-rag serve --mcp --stdio
```

## Intro

haiku.rag 是带 CLI、Python API 与 MCP server 的 agentic RAG 工具；已验证 524★，支持 `add-src`、`ask --cite` 与 `serve --mcp` 等命令。

**Best for:** 需要可引用、可复现，并偏好本地 LanceDB 的 RAG/Agent 团队

**Works with:** Python 3.12+，并需要配置 embedding provider（Ollama/OpenAI 等，README 说明）

**Setup time:** 6-15 minutes

### Key facts (verified)

- GitHub：524 stars · 35 forks；最近更新 2026-05-13。
- 许可证：MIT；作者头像与仓库链接均已通过 GitHub API 复核。
- README 中可对照的入口命令：`haiku-rag serve --mcp --stdio`。

## Main

- 先用 1 个 PDF 跑通并验证引用（`--cite`），再扩展到目录监控或 research agent 流程。

- 当你希望助手以“工具调用”而非粘贴上下文的方式工作时，启用 MCP server 模式（stdio）。

- 把 provider 选择写清楚：embedding 与 QA 模型可替换；为每次数据集运行记录 provider，保证可复现。

### Source-backed notes

- README 说明其基于 LanceDB、Pydantic AI 与 Docling，并同时提供 CLI 与 Python API 两种入口。
- README 给出 MCP server 用法：`haiku-rag serve --mcp --stdio`，并提供 `mcpServers` 的 JSON 配置示例。
- README 列举特性：混合检索、带页码/标题的引用，以及本地优先的嵌入式 LanceDB 存储等。

### FAQ

- **必须配置 embedding provider 吗？**：是的。README 说明索引/检索前需要配置（Ollama/OpenAI 等）。
- **能通过 MCP 客户端用吗？**：可以。运行 `serve --mcp --stdio` 并按示例加入客户端配置。
- **有精简版安装吗？**：有。README 提到 `haiku.rag-slim`，适合想减少依赖时使用。

## Source & Thanks

> Source: https://github.com/ggozad/haiku.rag
> License: MIT
> GitHub stars: 524 · forks: 35


---
Source: https://tokrepo.com/en/workflows/haiku-rag-agentic-rag-cli-mcp-server
Author: Agent Toolkit