About dbt Model Context Protocol (MCP) server
The dbt MCP server provides a standardized framework that lets you integrate AI applications with dbt‑managed data assets across different data platforms. This ensures consistent, governed access to models, metrics, lineage, and freshness across your AI tools.
To help with dbt, assistants need your project metadata and, when you allow it, supported actions such as CLI runs, platform APIs, and Semantic Layer queries. The dbt MCP server exposes those to MCP clients and supports use cases such as conversational access to data, agentic automation for dbt workflows, and AI-assisted development. This page covers local and remote setups, available tools, and how to get started.
The MCP server provides access to the dbt CLI, API, the Discovery API, and Semantic Layer. It provides access to private APIs, text-to-SQL, and SQL execution.
For more information on MCP, have a look at Get started with the Model Context Protocol (MCP).
The dbt MCP server comes in two flavors: local and remote.
- Local MCP server: runs locally on your machine and requires installing
uvx(which installs dbt-mcp locally). - Remote MCP server: uses an HTTP connection and makes calls to dbt-mcp hosted on the managed dbt platform. This setup requires no local installation and is ideal for data consumption use cases.
For more details on the server types, refer to Server access.
Get started
To get started, choose the quickstart that matches your setup:
To configure or disable specific tools (local MCP), see the Environment variables reference.
Server access
You can use the dbt MCP server in the following ways:
- Local MCP server — runs locally on your machine and requires installing
uvx(which installs dbt-mcp locally) and then runninguvx dbt-mcpto start the server. No need to clone the repo unless you want to contribute to dbt MCP server. - Remote MCP server — uses an HTTP connection and makes calls to dbt-mcp hosted on the managed dbt platform. This setup requires no local installation and is ideal for data consumption use cases.
Local MCP server
The local MCP server provides the best experience for development workflows, like authoring dbt models, tests, and documentation.
The local MCP server runs on your machine and requires installing uvx (which installs dbt-mcp locally) and then running uvx dbt-mcp to start the server. You don't need to clone the repository unless you want to contribute to dbt MCP. The local MCP server provides:
- Full access to dbt CLI commands (
dbt run,dbt build,dbt test, and more) - Support for dbt Core, dbt CLI, and dbt Fusion engine
- Ability to work with local dbt projects with or without a dbt platform account
- Optional integration with dbt platform APIs for metadata discovery and Semantic Layer access
Remote MCP server
The remote MCP server from dbt offers data consumption use cases without local setup. It doesn't support local development or dbt CLI commands; use the local MCP server for those workflows.
The remote MCP server connects to the dbt platform via HTTP and requires no local installation. This option is useful when:
- You either don’t want to install, or are restricted from installing, additional software on your system.
- Your use case is primarily consumption-based (for example, querying metrics, exploring metadata, viewing lineage).
Only text_to_sql consumes dbt Copilot credits. Other MCP tools do not.
When your account runs out of Copilot credits, the remote MCP server blocks all tools that run through it, even tools invoked from a local MCP server and proxied to remote MCP (like SQL and remote Fusion tools).
If you reach your dbt Copilot usage limit, all tools will be blocked until your Copilot credits reset. If you need help, please reach out to your account manager.
Supported tools by MCP server type
The dbt MCP server has access to many parts of the dbt experience related to development, deployment, and discovery. Here are the categories of tools supported based on what form of the MCP server you connect to as well as detailed information on exact commands or queries available to the LLM.
Local MCP is required for dbt CLI commands, Codegen, and Administrative API; remote MCP supports Semantic Layer, SQL, Discovery, and Fusion tools only.
Note that access to the dbt APIs is limited depending on your plan type.
| Loading table... |
Available tools
The dbt MCP server has access to many tools related to development, deployment, and discovery — like CLI
A full list of tools is available for your MCP server and is auto-fetched from the dbt MCP server README on GitHub when the docs are built, so it stays in sync with each release.
To view the full list of tools, see Available tools.
MCP integrations
The dbt MCP server integrates with any MCP client that supports token authentication and tool use capabilities. We have created integration guides for the following clients:
Resources
- Environment variables reference — full list of variables and tool configuration for local MCP
- For more information, refer to our blog on Introducing the dbt MCP Server.
Was this page helpful?
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
