dbt MCP Server

What is dbt MCP Server?

The dbt MCP Server is a bridge between dbt and AI-powered interfaces, exposing dbt’s core functionality, metadata, and analytical semantics as standardized tools that can be called by LLMs and applications via the MCP. This enables users to automate, query, and manage their dbt workflows and data metrics from natural language interfaces and LLM-driven agents with fine-grained access control and consistent protocol-based integration.

How to Configure dbt MCP Server

Install the server via the provided script:
```
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/dbt-labs/dbt-mcp/refs/heads/main/install.sh)"
```
This installs or updates dbt MCP Server in your home directory under ~/.dbt-mcp/.
Set environment variables for required settings such as DBT_HOST, DBT_TOKEN, and project-specific paths (DBT_PROJECT_DIR, etc.). You can customize which tool groups (CLI, Semantic Layer, Discovery API) are enabled via flags like DISABLE_DBT_CLI=false, or configure semantic/remote features as needed.
Configure the MCP client (e.g., Claude Desktop, Cursor, VS Code) by specifying the server's command, arguments, and environment configuration. Add the server details to the respective MCP config files (claude_desktop_config.json, mcp.json, settings.json), updating the executable path for your OS.

How to Use dbt MCP Server

Connect your MCP Client (such as Claude Desktop, Cursor, or VS Code) to the dbt MCP Server by adding its configuration in your client settings.
Start the MCP server using the installed command with the correct arguments. The server now exposes dbt tools to your connected clients.
From the client, invoke dbt features using natural language requests or through UI commands; the client will translate these to MCP tool calls, which are then executed by the server against your dbt project.
Manage, monitor, or stop the server from your client's MCP interface or via the command line.
For CLI-based usage, ensure correct environments and dependencies are set, and always review which tool groups are active.

Key Features

Seamless, protocol-based access to dbt’s core, semantic, and discovery APIs directly from AI interfaces
Secure execution of dbt CLI commands, semantic metric retrieval, and project metadata queries
Fine-grained configuration and enablement of tool groups (CLI, Semantic Layer, Discovery)
Integration-ready with modern LLM clients (Claude Desktop, Cursor, VS Code, and more)
Lower friction for LLM agents to perform both read-only and state-altering dbt operations
Easy installation, unified configuration, and automatic update support

Use Cases

Enable LLM-based copilots in code editors (like VS Code or Cursor) to automate dbt workflows, such as building or testing models
Let natural language agents fetch dbt project documentation or metadata for developer assistance or onboarding
Power AI dashboards or chatbots with access to semantic metrics via dbt’s Semantic Layer, enabling conversational analytics
Integrate dbt operational checkpoints (build/test/run) into automated pipelines orchestrated by LLM-driven agents
Facilitate ad-hoc querying, model exploration, or root cause analysis from conversational UIs using dbt discovery APIs

FAQ

Q: Do I need to deploy dbt MCP Server on the same machine as my dbt project?
A: The server can be run locally or remotely, but it will require access to the relevant project files and environment. You can specify paths and environment variables to ensure it locates your dbt project correctly.

Q: Is it safe to allow LLMs to access all dbt tools via MCP?
A: Use caution; some dbt tools (like build, run, and test) can alter your warehouse or models. Only enable tools you fully trust your clients to use, and prefer read-only modes (e.g., Discovery and Semantic Layer) where appropriate.

Q: How do I update the server if new versions are released?
A: Simply rerun the installation script; it will detect and update your existing dbt-mcp installation safely.

Q: Which dbt versions are supported?
A: dbt MCP Server is compatible with both dbt Core and dbt Cloud environments. Make sure your CLI path and dbt version match what is specified in your configuration.

Q: Can I selectively enable or disable certain tool groups?
A: Yes; use environment variables like DISABLE_DBT_CLI or DISABLE_DISCOVERY to turn features on or off at server startup.

Tool Name	Description
build	Executes dbt models, tests, snapshots, and seeds in dependency order within your project, materializing results in the warehouse.
compile	Compiles and validates models, tests, and analyses, outputting the generated SQL without running them.
docs	Builds and serves documentation for your dbt project, allowing interactive exploration of models, sources, and lineage.
ls	Lists dbt resources such as models, seeds, snapshots, and tests with filtering options for finer project exploration.
parse	Parses and validates dbt project files for syntax and logic correctness, helping catch errors early.
run	Executes and materializes selected dbt models to your data warehouse as per current configuration.
test	Runs dbt tests to validate model integrity and data quality rules defined in your project.
show	Executes an interactive ad-hoc query against your data warehouse from within the dbt context.
list_metrics	Retrieves all defined semantic metrics from your dbt Semantic Layer for further querying or inspection.
get_dimensions	Returns the dimensions associated with specified metrics from the Semantic Layer.
get_entities	Retrieves entities (e.g., primary keys or business objects) associated with specific metrics.
query_metrics	Allows querying of semantic metrics with support for grouping, filtering, ordering, and limiting results.
get_mart_models	Lists all dbt mart models present in your project, useful for analysts or data cataloging.
get_all_models	Returns all dbt models defined within your project, including their basic metadata.
get_model_details	Fetches detailed information about a given dbt model, including schema, lineage, and dependencies.
get_model_parents	Returns the upstream parent models for a specified model, aiding in dependency analysis or project navigation.