Creativity · MCP — server

MCP dbt Server

dbt is the de-facto transformation layer in the modern data stack. The MCP dbt Server exposes project metadata — models, tests, sources, exposures — and dbt CLI actions to MCP clients. Claude can then help author new models, run dbt build, and explain lineage from the graph.

MCP facts

Kind: server
Ecosystem: anthropic-mcp
Language: Python
Transports: stdio

Capabilities

Tools: list_models, describe_model, run_model, test_model, compile_sql
Resources: dbt:// URIs for each model's compiled SQL
Supports dbt Core locally and dbt Cloud via API token

Install

pip install mcp-dbt

Configuration

{
  "mcpServers": {
    "dbt": {
      "command": "uvx",
      "args": ["mcp-dbt", "--project-dir", "/Users/you/dbt-project"],
      "env": {
        "DBT_PROFILES_DIR": "/Users/you/.dbt"
      }
    }
  }
}

Frequently asked questions

What can I actually ask Claude with this?

"Which models depend on fct_orders?", "Run dbt test for the staging layer", or "Draft a new mart on top of stg_users". The server turns the dbt DAG into tool calls.

Does it support dbt Cloud?

Yes — configure DBT_CLOUD_API_TOKEN and DBT_CLOUD_ACCOUNT_ID to trigger jobs and fetch run artifacts.

Is it safe to run dbt from an LLM?

Treat it like running dbt in CI: sandbox the warehouse role, dry-run with compile, and gate run/build behind approval in your MCP client.

Sources

dbt Documentation — accessed 2026-04-20
Model Context Protocol — accessed 2026-04-20