dbt-core

dbt-core-mcp provides a language-driven interface to your dbt projects without requiring you to install or manage a dbt-core installation locally. It adapts to your environment, detects your dbt version and adapter, and lets Copilot execute dbt commands, run tests, and analyze lineage and impact directly in your workspace. You can query project metadata, view models, sources, and tests, and ask for downstream impact or compilation details. The server supports natural language commands like running only changed models, testing downstream, or inspecting specific models or CTEs, and it returns concrete results rather than just suggestions. To get started, connect your MCP client to the dbt-core-mcp server as configured, then use prompts to trigger dbt runs, queries, and analyses against your actual environment and project structure.

Prerequisites:\n- Python 3.9+ installed on your system.\n- Access to a compatible MCP client (e.g., VS Code MCP extension).\n- Optional: uv (Python package) if you want to run the server via uvx.\n\nStep 1: Install uv (if not already installed)\nbash\npip install uv\n``\nStep 2: Install the MCP server (dbt-core-mcp) in your environment via uvx (preferred)\nbash\n# Ensure uv is installed, then install the server via uvx using the provided identifier\nuvx install dbt-core-mcp\n``\nStep 3: Configure your MCP client to point to the server\n- One-click installation routes in the README links configure uvx with an argument like: dbt-core-mcp\n- Or manually create your .vscode/mcp.json with the following (example):\njson\n{\n "servers": {\n "dbt-core": {\n "command": "uvx",\n "args": ["dbt-core-mcp"]\n }\n }\n}\n\nStep 4: (Optional) Update to latest features from GitHub\nbash\nuv cache clean dbt-core-mcp\n# Restart your MCP client to pick up changes\n

Tips and common issues:\n- The server does not require a local dbt-core installation; it executes dbt commands using your workspace's environment. Ensure your virtual environment and adapters (e.g., postgres, snowflake) are accessible from the runtime you are using.\n- If you encounter caching issues with git installations (uvx or pipx), use the provided cache-clean steps in the bleeding-edge configuration notes and restart the MCP client.\n- You can use command timeouts to limit long-running dbt commands; by default there is no timeout. In uvx, you can pass --dbt-command-timeout as shown in the optional configuration snippet.\n- The server supports any dbt adapter and works with your existing dbt setup; you can run only changed models, downstream tests, and inspect lineage or compiled SQL.\n- If you switch between environments (venv, poetry, conda), ensure the active Python and environment provide the correct dbt configuration and adapters for your project.