# Ragstar — dbt Knowledge Base + OAuth MCP Server

> Ragstar is a self-hosted dbt knowledge base with a web UI and an OAuth-protected MCP server so agents can search models, lineage, and docs safely.

## Install

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

## Quick Use

1. Clone and start the stack:
   ```bash
   git clone https://github.com/pragunbhutani/ragstar.git && cd ragstar
   cp .env.example .env
   docker compose up --build -d
   ```
2. Run first-time migrations:
   ```bash
   docker compose exec backend-django uv run python manage.py migrate
   ```
3. Verify:
   - Web UI: `http://localhost:3000`
   - MCP health: `curl http://localhost:8080/health`

## Intro

Ragstar is a self-hosted dbt knowledge base with a web UI and an OAuth-protected MCP server so agents can search models, lineage, and docs safely.

- **Best for:** data teams using dbt who want an LLM-facing knowledge base with tools (not just exported docs)
- **Works with:** Docker Compose, Postgres + Redis in the stack, dbt project docs/metadata, MCP clients for the built-in server
- **Setup time:** 25-60 minutes

## Practical Notes

- Quant: the stack exposes multiple services (UI, API, worker, MCP) so budget time for ports and env configuration.
- Quant: the MCP server provides tools such as listing models, semantic search, and retrieving model details including lineage.

## Why it matters

Ragstar is notable because it treats dbt assets as an operational knowledge base with an MCP tool boundary and OAuth, rather than a static docs export.

- Concrete two-command quick start plus clear service URLs makes onboarding straightforward.
- The MCP server is designed for self-hosted deployment with an OAuth flow, fitting enterprise access expectations.
- Tool list focuses on practical dbt workflows: browse, search, and inspect model details and lineage.

## Rollout pattern

- Connect one dbt project first and validate that search results map to real model owners and descriptions.
- Decide which environments get MCP access; keep production knowledge bases separated from experiments.
- Add an internal policy: agents may retrieve model details, but changes to dbt code still go through PR review.

## Watchouts

Because Ragstar includes an MCP server and a web UI, treat it as a real internal service: OAuth configured, secrets managed, and environments separated.

### FAQ

**Q: Is the MCP server available in hosted mode?**
A: The README states MCP is only for self-hosted/open source deployments due to a 1:1 client-server relationship.

**Q: What can the MCP server do?**
A: It can list dbt models, search semantically, and fetch model details and project summaries.

**Q: What is the safest first use case?**
A: Use it as read-only discovery: find the right model and lineage, then do changes via PRs in your dbt repo.

## Source & Thanks

> Source: https://github.com/pragunbhutani/dbt-llm-agent
> License: MIT
> GitHub stars: 171 · forks: 32

---

<!-- ZH -->

## 快速使用

1. 克隆并启动：
   ```bash
   git clone https://github.com/pragunbhutani/ragstar.git && cd ragstar
   cp .env.example .env
   docker compose up --build -d
   ```
2. 首次迁移：
   ```bash
   docker compose exec backend-django uv run python manage.py migrate
   ```
3. 验证：
   - Web UI：`http://localhost:3000`
   - MCP 健康检查：`curl http://localhost:8080/health`

## 简介

Ragstar 是自托管的 dbt 知识库与可视化界面，并内置 OAuth 保护的 MCP Server，让 Agent 能检索模型、血缘与文档而无需手动导出，适合数据团队把 dbt 资产变成可问可查的工具层。

- **适合谁：** 使用 dbt 的数据团队：希望有面向 LLM 的“可调用工具”知识库，而不是只导出文档
- **可搭配：** Docker Compose（含 Postgres/Redis）、dbt 项目元数据/文档、可连接内置 MCP 的客户端
- **准备时间：** 25-60 分钟

## 实战建议

- 量化信息：栈里包含 UI/API/worker/MCP 多个服务，端口与 env 配置需要预留时间。
- 量化信息：MCP server 提供列模型、语义检索、获取模型详情（含血缘）等工具。

## 为什么值得收录

Ragstar 的亮点在于：它把 dbt 资产做成“可操作的知识库”，并通过 MCP + OAuth 给工具调用划出明确边界，而不是停留在静态文档导出。

- 两条命令即可启动，并清楚列出各服务 URL，上手快。
- MCP server 面向自托管并实现 OAuth 流程，更符合企业内部访问预期。
- 工具列表聚焦 dbt 的真实需求：浏览、检索、查看模型详情与血缘。

## 落地路径

- 先接入 1 个 dbt 项目，验证检索结果与真实 owner/描述一致。
- 明确哪些环境开放 MCP；生产知识库与实验环境要隔离。
- 制定内部规则：Agent 只做只读发现，任何 dbt 代码变更走 PR review。

## 注意事项

Ragstar 同时包含 MCP server 与 Web UI，要当作内部服务治理：OAuth 配好、secret 管好、环境隔离好。

### FAQ

**托管版也能用 MCP 吗？**
答：README 表示 MCP 仅支持自托管/开源部署（客户端与服务端 1:1 关系）。

**MCP server 能做什么？**
答：可列出模型、语义检索、获取模型详情/血缘与项目概览。

**最安全的起步场景？**
答：把它当作只读发现工具：找对模型与血缘；修改仍在 dbt 仓库里走 PR。

## 来源与感谢

> Source: https://github.com/pragunbhutani/dbt-llm-agent
> License: MIT
> GitHub stars: 171 · forks: 32


---
Source: https://tokrepo.com/en/workflows/ragstar-dbt-knowledge-base-oauth-mcp-server
Author: MCP Hub