databricks-mcp

Instructions

Implementation Reference

• tools/query.py:10-36 (handler)

  The databricks_query function is defined and decorated as an MCP tool, containing logic to validate against destructive SQL operations before calling execute_sql.

  @mcp.tool
  def databricks_query(ctx: Context, sql_query: Optional[str] = None, sql: Optional[str] = None) -> List[Dict[str, Any]]:
      """
      Execute Databricks SQL query (supports SELECT, SHOW, DESCRIBE, CREATE, ALTER).
      INSERT, UPDATE, DELETE, DROP and other destructive operations are blocked.
  
      Args:
          sql_query: SQL query statement (preferred parameter)
          sql: SQL query statement (fallback for backward compatibility)
  
      Returns:
          Query results as list of dicts
      """
      query_to_execute = sql_query if sql_query is not None else sql
  
      if query_to_execute is None:
          raise ToolError("Must provide sql_query or sql parameter.")
  
      query_upper = query_to_execute.strip().upper()
      forbidden_keywords = [
          "INSERT INTO", "UPDATE ", "DELETE FROM", "DROP TABLE", "DROP VIEW",
          "DROP SCHEMA", "DROP CATALOG", "TRUNCATE TABLE", "MERGE INTO", "COPY INTO"
      ]
      for keyword in forbidden_keywords:
          if keyword in query_upper:
              raise ToolError(f"Destructive operation not allowed: {keyword}")
      return execute_sql(ctx, query_to_execute)