DBT Core MCP Server

Instructions

Implementation Reference

• src/dbt_core_mcp/server.py:1107-1187 (handler)

  The core handler function that implements the seed_data tool logic. It constructs the dbt seed command with selectors (manual or state-based), executes it via the runner, handles progress reporting, parses results, and saves state for future runs.

  async def toolImpl_seed_data(
      self,
      ctx: Context | None = None,
      select: str | None = None,
      exclude: str | None = None,
      select_state_modified: bool = False,
      select_state_modified_plus_downstream: bool = False,
      full_refresh: bool = False,
      show: bool = False,
  ) -> dict[str, Any]:
      """Implementation of seed_data tool."""
      # Prepare state-based selection (validates and returns selector)
      selector = await self._prepare_state_based_selection(select_state_modified, select_state_modified_plus_downstream, select)
  
      # Early return if state-based requested but no state exists
      if select_state_modified and not selector:
          return {
              "status": "success",
              "message": "No previous state found - cannot determine modifications",
              "results": [],
              "elapsed_time": 0,
          }
  
      # Build command args
      args = ["seed"]
  
      # Add selector if we have one (state-based or manual)
      if selector:
          args.extend(["-s", selector, "--state", "target/state_last_run"])
      elif select:
          args.extend(["-s", select])
  
      if exclude:
          args.extend(["--exclude", exclude])
  
      if full_refresh:
          args.append("--full-refresh")
  
      if show:
          args.append("--show")
  
      # Execute with progress reporting
      logger.info(f"Running DBT seed with args: {args}")
  
      # Define progress callback if context available
      async def progress_callback(current: int, total: int, message: str) -> None:
          if ctx:
              await ctx.report_progress(progress=current, total=total, message=message)
  
      result = await self.runner.invoke(args, progress_callback=progress_callback if ctx else None)  # type: ignore
  
      if not result.success:
          error_msg = str(result.exception) if result.exception else "Seed failed"
          response = {
              "status": "error",
              "message": error_msg,
              "command": " ".join(args),
          }
          # Include dbt output for debugging
          if result.stdout:
              response["dbt_output"] = result.stdout
          if result.stderr:
              response["stderr"] = result.stderr
          return response
  
      # Save state on success for next modified run
      if result.success and self.project_dir:
          state_dir = self.project_dir / "target" / "state_last_run"
          state_dir.mkdir(parents=True, exist_ok=True)
          manifest_path = self.runner.get_manifest_path()  # type: ignore
          shutil.copy(manifest_path, state_dir / "manifest.json")
  
      # Parse run_results.json for details
      run_results = self._parse_run_results()
  
      return {
          "status": "success",
          "command": " ".join(args),
          "results": run_results.get("results", []),
          "elapsed_time": run_results.get("elapsed_time"),
      }

• src/dbt_core_mcp/server.py:1639-1684 (registration)

  The FastMCP tool registration decorator (@self.app.tool()) that defines the 'seed_data' tool, its parameters (serving as input schema), comprehensive docstring, and delegates to the implementation handler.

  async def seed_data(
      ctx: Context,
      select: str | None = None,
      exclude: str | None = None,
      select_state_modified: bool = False,
      select_state_modified_plus_downstream: bool = False,
      full_refresh: bool = False,
      show: bool = False,
  ) -> dict[str, Any]:
      """Load seed data (CSV files) from seeds/ directory into database tables.
  
      Seeds are typically used for reference data like country codes, product categories, etc.
  
      State-based selection modes (detects changed CSV files):
      - select_state_modified: Load only seeds modified since last successful run (state:modified)
      - select_state_modified_plus_downstream: Load modified + downstream dependencies (state:modified+)
        Note: Requires select_state_modified=True
  
      Manual selection (alternative to state-based):
      - select: dbt selector syntax (e.g., "raw_customers", "tag:lookup")
      - exclude: Exclude specific seeds
  
      Important: Change detection for seeds works via file hash comparison:
      - Seeds < 1 MiB: Content hash is compared (recommended)
      - Seeds >= 1 MiB: Only file path changes are detected (content changes ignored)
      For large seeds, use manual selection or run all seeds.
  
      Args:
          select: Manual selector for seeds
          exclude: Exclude selector
          select_state_modified: Use state:modified selector (changed seeds only)
          select_state_modified_plus_downstream: Extend to state:modified+ (changed + downstream)
          full_refresh: Truncate and reload seed tables (default behavior)
          show: Show preview of loaded data
  
      Returns:
          Seed results with status and loaded seed info
  
      Examples:
          seed_data()  # Load all seeds
          seed_data(select_state_modified=True)  # Load only changed CSVs
          seed_data(select="raw_customers")  # Load specific seed
      """
      await self._ensure_initialized_with_context(ctx)
      return await self.toolImpl_seed_data(ctx, select, exclude, select_state_modified, select_state_modified_plus_downstream, full_refresh, show)

• src/dbt_core_mcp/server.py:479-520 (helper)

  Helper method used by seed_data (and similar tools) to validate parameters and construct state-based dbt selectors like 'state:modified' or 'state:modified+'.

  async def _prepare_state_based_selection(
      self,
      select_state_modified: bool,
      select_state_modified_plus_downstream: bool,
      select: str | None,
  ) -> str | None:
      """Validate and prepare state-based selection.
  
      Args:
          select_state_modified: Use state:modified selector
          select_state_modified_plus_downstream: Extend to state:modified+
          select: Manual selector (conflicts with state-based)
  
      Returns:
          The dbt selector string to use ("state:modified" or "state:modified+"), or None if:
          - Not using state-based selection
          - No previous state exists (cannot determine modifications)
  
      Raises:
          ValueError: If validation fails
      """
      # Validate: hierarchical requirement
      if select_state_modified_plus_downstream and not select_state_modified:
          raise ValueError("select_state_modified_plus_downstream requires select_state_modified=True")
  
      # Validate: can't use both state-based and manual selection
      if select_state_modified and select:
          raise ValueError("Cannot use both select_state_modified* flags and select parameter")
  
      # If not using state-based selection, return None
      if not select_state_modified:
          return None
  
      # Check if state exists
      state_dir = self.project_dir / "target" / "state_last_run"  # type: ignore
      if not state_dir.exists():
          # No state - cannot determine modifications
          return None
  
      # Return selector (state exists)
      return "state:modified+" if select_state_modified_plus_downstream else "state:modified"

• src/dbt_core_mcp/server.py:304-341 (helper)

  Helper method to parse dbt's run_results.json file after seed/run/test/build operations, simplifying results for tool responses.

  def _parse_run_results(self) -> dict[str, Any]:
      """Parse target/run_results.json after dbt run/test/build.
  
      Returns:
          Dictionary with results array and metadata
      """
      if not self.project_dir:
          return {"results": [], "elapsed_time": 0}
  
      run_results_path = self.project_dir / "target" / "run_results.json"
      if not run_results_path.exists():
          return {"results": [], "elapsed_time": 0}
  
      try:
          with open(run_results_path) as f:
              data = json.load(f)
  
          # Simplify results for output
          simplified_results = []
          for result in data.get("results", []):
              simplified_results.append(
                  {
                      "unique_id": result.get("unique_id"),
                      "status": result.get("status"),
                      "message": result.get("message"),
                      "execution_time": result.get("execution_time"),
                      "failures": result.get("failures"),
                  }
              )
  
          return {
              "results": simplified_results,
              "elapsed_time": data.get("elapsed_time", 0),
          }
      except Exception as e:
          logger.warning(f"Failed to parse run_results.json: {e}")
          return {"results": [], "elapsed_time": 0}