DBT CLI MCP Server

The DBT CLI MCP Server bridges the gap between AI coding assistants and data transformation workflows by exposing the full suite of dbt CLI commands as standardized MCP tools. Developers who rely on AI assistants for data engineering tasks can now invoke dbt operations—such as running models, compiling SQL, testing schemas, or seeding data—directly from the assistant’s interface without leaving the conversational context. This eliminates manual terminal work and streamlines end‑to‑end data pipeline management.

At its core, the server wraps every major dbt operation in a lightweight tool: , , , , and more. Each tool accepts the same set of arguments that a normal dbt command would, including model selectors, project directories, and profile configurations. The server also manages environment variables and logging levels, allowing a single configuration to control the entire dbt ecosystem. By exposing these commands through MCP, an AI assistant can ask a user to “run the customer models” or “list all incremental tables,” and the assistant will translate that request into a precise dbt CLI call, returning structured results back to the user.

Key capabilities include:

• Absolute project path enforcement – ensures that all operations target the correct dbt project, preventing accidental runs in wrong directories.
• Configurable executable and profile paths – developers can point the server to any dbt binary or profiles.yml location, supporting custom installations and multi‑environment setups.
• Environment file integration – a single file can inject credentials and variables, keeping secrets out of the command line.
• Full resource enumeration – lists models, sources, seeds, and snapshots in JSON format, enabling assistants to provide quick overviews or dependency graphs.

Real‑world use cases span from continuous integration pipelines, where an AI assistant triggers a after code review, to interactive data exploration sessions, where the assistant compiles a model and shows preview results with . In data‑driven product teams, an AI assistant can automatically seed test data with before a feature rollout, or compile and validate transformations on demand during stakeholder meetings.

Integration is straightforward: the server runs as an MCP process, and any client—such as Claude for Desktop or a custom web UI—can add it to its configuration. Once registered, the assistant can invoke any dbt command through a single tool call, receive structured JSON responses, and even chain multiple operations (e.g., compile then run). This tight coupling of AI intent with deterministic dbt execution eliminates friction, reduces errors, and accelerates the data engineering workflow.