fix(mcp): always filter list responses by columns_requested (#37505)

superset/mcp_service/chart/tool/list_charts.py

@@ -137,20 +137,17 @@ async def list_charts(request: ListChartsRequest, ctx: Context) -> ChartList:
             % (count, total_pages)
         )
 
-        # Apply field filtering via serialization context if select_columns specified
+        # Apply field filtering via serialization context
+        # Always use columns_requested (either explicit select_columns or defaults)
         # This triggers ChartInfo._filter_fields_by_context for each chart
-        if request.select_columns:
-            await ctx.debug(
-                "Applying field filtering via serialization context: select_columns=%s"
-                % (request.select_columns,)
-            )
-            # Return dict with context - FastMCP handles serialization
-            return result.model_dump(
-                mode="json", context={"select_columns": request.select_columns}
-            )
-
-        # No filtering - return full result as dict
-        return result.model_dump(mode="json")
+        columns_to_filter = result.columns_requested
+        await ctx.debug(
+            "Applying field filtering via serialization context: columns=%s"
+            % (columns_to_filter,)
+        )
+        return result.model_dump(
+            mode="json", context={"select_columns": columns_to_filter}
+        )
     except Exception as e:
         await ctx.error("Failed to list charts: %s" % (str(e),))
         raise

superset/mcp_service/dashboard/tool/list_dashboards.py

@@ -139,17 +139,12 @@ async def list_dashboards(
         % (count, total_pages)
     )
 
-    # Apply field filtering via serialization context if select_columns specified
+    # Apply field filtering via serialization context
+    # Always use columns_requested (either explicit select_columns or defaults)
     # This triggers DashboardInfo._filter_fields_by_context for each dashboard
-    if request.select_columns:
-        await ctx.debug(
-            "Applying field filtering via serialization context: select_columns=%s"
-            % (request.select_columns,)
-        )
-        # Return dict with context - FastMCP handles serialization
-        return result.model_dump(
-            mode="json", context={"select_columns": request.select_columns}
-        )
-
-    # No filtering - return full result as dict
-    return result.model_dump(mode="json")
+    columns_to_filter = result.columns_requested
+    await ctx.debug(
+        "Applying field filtering via serialization context: columns=%s"
+        % (columns_to_filter,)
+    )
+    return result.model_dump(mode="json", context={"select_columns": columns_to_filter})

superset/mcp_service/dataset/tool/list_datasets.py

@@ -148,20 +148,17 @@ async def list_datasets(request: ListDatasetsRequest, ctx: Context) -> DatasetLi
             )
         )
 
-        # Apply field filtering via serialization context if select_columns specified
+        # Apply field filtering via serialization context
+        # Always use columns_requested (either explicit select_columns or defaults)
         # This triggers DatasetInfo._filter_fields_by_context for each dataset
-        if request.select_columns:
-            await ctx.debug(
-                "Applying field filtering via serialization context: select_columns=%s"
-                % (request.select_columns,)
-            )
-            # Return dict with context - FastMCP handles serialization
-            return result.model_dump(
-                mode="json", context={"select_columns": request.select_columns}
-            )
-
-        # No filtering - return full result as dict
-        return result.model_dump(mode="json")
+        columns_to_filter = result.columns_requested
+        await ctx.debug(
+            "Applying field filtering via serialization context: columns=%s"
+            % (columns_to_filter,)
+        )
+        return result.model_dump(
+            mode="json", context={"select_columns": columns_to_filter}
+        )
 
     except Exception as e:
         await ctx.error(

tests/unit_tests/mcp_service/dataset/tool/test_dataset_tools.py

@@ -1239,6 +1239,59 @@ class TestDatasetDefaultColumnFiltering:
             assert "metrics" in data["columns_available"]
             assert "description" in data["columns_available"]
 
+    @patch("superset.daos.dataset.DatasetDAO.list")
+    @pytest.mark.asyncio
+    async def test_default_columns_filters_actual_response_data(
+        self, mock_list, mcp_server
+    ):
+        """Test that actual dataset items only contain default columns.
+
+        This verifies the fix for the bug where all 40+ columns were returned
+        with null values even when only default columns were requested.
+        See: https://github.com/apache/superset/pull/37213
+        """
+        dataset = create_mock_dataset()
+        mock_list.return_value = ([dataset], 1)
+
+        async with Client(mcp_server) as client:
+            # Empty select_columns = use default columns
+            request = ListDatasetsRequest(page=1, page_size=10, select_columns=[])
+            result = await client.call_tool(
+                "list_datasets", {"request": request.model_dump()}
+            )
+            data = json.loads(result.content[0].text)
+
+            # Get the actual dataset item
+            assert len(data["datasets"]) == 1
+            dataset_item = data["datasets"][0]
+
+            # Verify ONLY default columns are present in the response item
+            expected_keys = {"id", "table_name", "schema_name", "uuid"}
+            actual_keys = set(dataset_item.keys())
+
+            # The response should only contain the default columns, NOT all columns
+            # with null values (which was the bug)
+            assert actual_keys == expected_keys, (
+                f"Expected only default columns {expected_keys}, "
+                f"but got {actual_keys}. "
+                f"Extra columns: {actual_keys - expected_keys}"
+            )
+
+            # Verify non-default columns are NOT present (not even with null values)
+            non_default_columns = [
+                "description",
+                "database_name",
+                "changed_by",
+                "changed_on",
+                "columns",
+                "metrics",
+            ]
+            for col in non_default_columns:
+                assert col not in dataset_item, (
+                    f"Non-default column '{col}' should not be in response. "
+                    f"This indicates the column filtering is not working."
+                )
+
 
 class TestDatasetSortableColumns:
     """Test sortable columns configuration for dataset tools."""