# dbt-mcp — dbt Context MCP Server for Agents

> Give AI agents structured access to dbt project context and tools (SQL, semantic layer, docs search). Ships an experimental MCP bundle in releases.

## Install

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

## Quick Use

1. Install / run:
   ```bash
   gh release download -R dbt-labs/dbt-mcp -p dbt-mcp.mcpb
   ```
2. Start / smoke test:
   ```bash
   Import the `dbt-mcp.mcpb` bundle into an MCPB-aware client, then connect it to your dbt environment.
   ```
3. Verify:
   - Call a read-only tool first (e.g., list/search) and confirm it returns your project resources before enabling CLI tools.

## Intro

Give AI agents structured access to dbt project context and tools (SQL, semantic layer, docs search). Ships an experimental MCP bundle in releases.

- **Best for:** Analytics engineering teams who want an agent to answer dbt questions with real project context, not guesswork
- **Works with:** dbt Core/Fusion/Platform workflows + MCP clients (per dbt docs)
- **Setup time:** 15 minutes

## How to Use It Well

- Setup time ~15 minutes (download bundle + wire credentials + connect)
- GitHub stars + forks (verified): see Source & Thanks
- Tools span SQL, semantic layer queries, and docs search (see README tool list)

When agents answer data questions, the failure mode is hallucinating models, columns, or semantics. dbt-mcp is a way to ground answers in your dbt project artifacts and (optionally) your dbt platform APIs—so the agent can cite real entities.

### FAQ

**Q: Is it read-only?**
A: It includes both read-only context tools and CLI-like operations. Start with read-only and expand intentionally.

**Q: Do I need dbt Cloud/Platform?**
A: The README mentions supporting multiple dbt environments; confirm which mode you run and configure accordingly.

**Q: How do I keep it safe?**
A: Use least-privilege credentials, separate dev/prod targets, and require human confirmation for any write-like operation.

## Source & Thanks

> Source: https://github.com/dbt-labs/dbt-mcp
> License: Apache-2.0
> GitHub stars: 559 · forks: 117

---

<!-- ZH -->

## 快速使用

1. 安装 / 运行：
   ```bash
   gh release download -R dbt-labs/dbt-mcp -p dbt-mcp.mcpb
   ```
2. 启动 / 冒烟测试：
   ```bash
   Import the `dbt-mcp.mcpb` bundle into an MCPB-aware client, then connect it to your dbt environment.
   ```
3. 验证：
   - Call a read-only tool first (e.g., list/search) and confirm it returns your project resources before enabling CLI tools.

## 简介

为 AI agent 提供结构化的 dbt 项目上下文与工具能力（SQL、语义层、文档搜索等），并在 release 中提供 `dbt-mcp.mcpb` bundle 便于导入，减少手工配置。

- **适合谁：** 希望 agent 基于真实 dbt 项目上下文回答问题、而不是瞎猜的分析工程团队
- **可搭配：** dbt Core/Fusion/Platform 工作流 + MCP 客户端（以 dbt 文档为准）
- **准备时间：** 15 分钟

## 实战建议

- 接入约 15 分钟（下载 bundle + 配凭据 + 连接）
- GitHub stars + forks（已核验）：见「来源与感谢」
- 工具覆盖 SQL、语义层查询、文档搜索（见 README 工具列表）

Agent 做数据相关问答时，最常见的失败模式是“编模型/编字段/编语义”。dbt-mcp 的价值在于把回答锚定在真实的 dbt 项目产物上，并（可选地）连接到平台 API，让 agent 以真实实体为依据进行检索与解释。

### FAQ

**它是只读的吗？**
A: 既有只读上下文工具，也可能包含偏“CLI 操作”的能力。建议先只开只读，再按需逐步放开。

**必须用 dbt Cloud/Platform 吗？**
A: README 描述支持多种 dbt 环境；请按你实际运行方式配置并核对权限。

**怎么保证安全？**
A: 用最小权限凭据、区分 dev/prod 目标，并对任何可能改动资源的操作加人工确认。

## 来源与感谢

> Source: https://github.com/dbt-labs/dbt-mcp
> License: Apache-2.0
> GitHub stars: 559 · forks: 117


---
Source: https://tokrepo.com/en/workflows/dbt-mcp-dbt-context-mcp-server-for-agents
Author: MCP Hub