Metabase MCP Server

Instructions

Implementation Reference

• src/metabase_mcp_server.py:1310-1341 (handler)

  The main handler function for the execute_sql_query tool. It accepts database_id, query string, and optional parameters, constructs a native query payload, and sends it to Metabase's /api/dataset endpoint via the make_metabase_request helper.

  @mcp.tool()
  async def execute_sql_query(
      database_id: int,
      query: str,
      parameters: Optional[List] = None
  ) -> Dict[str, Any]:
      """
      Execute a native SQL query through Metabase.
  
      Args:
          database_id (int): ID of the database to execute the query on.
          native_query (str): The SQL query to execute.
          parameters (list, optional): Query parameters.
  
      Returns:
          Dict[str, Any]: Query execution result.
          
      Notes:
          - For PostgreSQL databases, column names are be case-sensitive
          - Use double quotes around column names with mixed case (e.g., "columnName")
          - Example with quoted column names: 
            SELECT "userId", "orderDate", COUNT(*) FROM "Orders" GROUP BY "userId", "orderDate"
      """
      query_payload = {
          "database": database_id,
          "type": "native",
          "native": {"query": query},
          "parameters": parameters or []
      }
      logger.info(f"Executing SQL query on database {database_id}")
      logger.debug(f"Query: {query[:100]}...")
      return await make_metabase_request(RequestMethod.POST, "/api/dataset", json=query_payload)

• src/metabase_mcp_server.py:159-160 (registration)

  Initializes the FastMCP server instance. The @mcp.tool() decorator on execute_sql_query registers it as an MCP tool with this server instance.

  # Initialize FastMCP agent
  mcp = FastMCP("metabase", lifespan=app_lifespan)

• src/metabase_mcp_server.py:162-247 (helper)

  Helper utility function that handles HTTP requests to the Metabase API. Used by execute_sql_query to post the SQL query to the /api/dataset endpoint. Includes error handling, logging, and response validation.

  async def make_metabase_request(
      method: RequestMethod,
      endpoint: str,
      data: Optional[Dict[str, Any] | bytes] = None,
      params: Optional[Dict[str, Any]] = None,
      json: Any = None,
      headers: Optional[Dict[str, str]] = None,
  ) -> Dict[str, Any]:
      """
      Make a request to the Metabase API.
      
      Args:
          method: HTTP method to use (GET, POST, PUT, DELETE)
          endpoint: API endpoint path
          data: Request data (for form data)
          params: URL parameters
          json: JSON request body
          headers: Additional headers
          
      Returns:
          Dict[str, Any]: Response data
          
      Raises:
          MetabaseConnectionError: When the Metabase server is unreachable
          MetabaseResponseError: When Metabase returns a non-2xx status code
          RuntimeError: For other errors
      """
      
      if not METABASE_URL or not METABASE_API_KEY:
          raise RuntimeError("METABASE_URL or METABASE_API_KEY environment variable is not set. Metabase API requests will fail.")
  
      if session is None:
          raise RuntimeError("HTTP session is not initialized. Ensure app_lifespan was called.")
  
      try:
          request_headers = headers or {}
          
          logger.debug(f"Making {method.name} request to {METABASE_URL}{endpoint}")
          
          # Log request payload for debugging (omit sensitive info)
          if json and logger.level <= logging.DEBUG:
              sanitized_json = {**json}
              if 'password' in sanitized_json:
                  sanitized_json['password'] = '********'
              logger.debug(f"Request payload: {sanitized_json}")
              
          response = await session.request(
              method=method.name,
              url=endpoint,
              timeout=aiohttp.ClientTimeout(total=30),
              headers=request_headers,
              data=data,
              params=params,
              json=json,
          )
  
          try:
              # Handle 500 errors with more detailed info
              if response.status >= 500:
                  error_text = await response.text()
                  logger.error(f"Server error {response.status}: {error_text[:200]}")
                  raise MetabaseResponseError(response.status, f"Server Error: {error_text[:200]}", endpoint)
              
              response.raise_for_status()
              response_data = await response.json()
              
              # Ensure the response is a dictionary for FastMCP compatibility
              return ensure_dict_response(response_data)
              
          except aiohttp.ContentTypeError:
              # Handle empty responses or non-JSON responses
              content = await response.text()
              if not content:
                  return {"data": {}}
              logger.warning(f"Received non-JSON response: {content}")
              return {"data": content}
  
      except aiohttp.ClientConnectionError as e:
          logger.error(f"Connection error: {str(e)}")
          raise MetabaseConnectionError("Metabase is unreachable. Is the Metabase server running?") from e
      except aiohttp.ClientResponseError as e:
          logger.error(f"Response error: {e.status}, {e.message}, {e.request_info.url}")
          raise MetabaseResponseError(e.status, e.message, str(e.request_info.url)) from e
      except Exception as e:
          logger.error(f"Request error: {str(e)}")
          raise RuntimeError(f"Request error: {str(e)}") from e