diff --git a/climetlab/readers/__init__.py b/climetlab/readers/__init__.py
index af52116b..16b89114 100644
--- a/climetlab/readers/__init__.py
+++ b/climetlab/readers/__init__.py
@@ -37,6 +37,8 @@ def multi_merge(cls, readers):
 
 
 class MultiReaders:
+    backend_kwargs = {}
+
     def __init__(self, readers):
         self.readers = readers
 
@@ -58,6 +60,8 @@ def preprocess(ds):
                 assert options == opts, f"{options} != {opts}"
 
         options.update(kwargs)
+        options.setdefault("backend_kwargs", {})
+        options["backend_kwargs"].update(self.backend_kwargs)
 
         return xr.open_mfdataset(
             [r.path for r in self.readers],
diff --git a/climetlab/readers/grib.py b/climetlab/readers/grib.py
index aa513ead..b3371563 100644
--- a/climetlab/readers/grib.py
+++ b/climetlab/readers/grib.py
@@ -220,6 +220,7 @@ def __iter__(self):
 
 class MultiGribReaders(MultiReaders):
     engine = "cfgrib"
+    backend_kwargs = {"squeeze": False}
 
 
 class GRIBReader(Reader):
@@ -249,12 +250,14 @@ def __getitem__(self, n):
     def __len__(self):
         return len(self._items())
 
-    def to_xarray(self):
-        import xarray as xr
+    def to_xarray(self, **kwargs):
+        # So we use the same code
+        return MultiGribReaders([self]).to_xarray(**kwargs)
+        # import xarray as xr
 
-        params = self.source.cfgrib_options()
-        ds = xr.open_dataset(self.path, engine="cfgrib", **params)
-        return self.source.post_xarray_open_dataset_hook(ds)
+        # params = self.source.cfgrib_options()
+        # ds = xr.open_dataset(self.path, engine="cfgrib", **params)
+        # return self.source.post_xarray_open_dataset_hook(ds)
 
     # @dict_args
     # def sel(self, **kwargs):
diff --git a/climetlab/readers/netcdf.py b/climetlab/readers/netcdf.py
index 41c629ba..eb85f050 100644
--- a/climetlab/readers/netcdf.py
+++ b/climetlab/readers/netcdf.py
@@ -312,8 +312,10 @@ def _get_fields(self, ds):  # noqa C901
 
         return fields
 
-    def to_xarray(self):
-        return xr.open_dataset(self.path, engine="netcdf4")
+    def to_xarray(self, **kwargs):
+        # So we use the same code
+        return MultiNetcdfReaders([self]).to_xarray(**kwargs)
+        # return xr.open_dataset(self.path, engine="netcdf4")
 
     @classmethod
     def multi_merge(cls, readers):
diff --git a/tests/test_grib.py b/tests/test_grib.py
index d70e1a53..09768243 100644
--- a/tests/test_grib.py
+++ b/tests/test_grib.py
@@ -34,7 +34,7 @@ def test_sel():
     s.sel(shortName="2t")
 
 
-@pytest.mark.skipif(("GITHUB_WORKFLOW" in os.environ) or True, reason="Not yet ready")
+# @pytest.mark.skipif(("GITHUB_WORKFLOW" in os.environ) or True, reason="Not yet ready")
 def test_multi():
     s1 = load_source(
         "cds",
diff --git a/tests/test_netcdf.py b/tests/test_netcdf.py
index 4081fe4b..1b2b15da 100644
--- a/tests/test_netcdf.py
+++ b/tests/test_netcdf.py
@@ -26,6 +26,7 @@ def test_multi():
         date="2021-03-01",
         format="netcdf",
     )
+    print(s1.to_xarray())
     s2 = load_source(
         "cds",
         "reanalysis-era5-single-levels",
@@ -34,6 +35,7 @@ def test_multi():
         date="2021-03-02",
         format="netcdf",
     )
+    print(s2.to_xarray())
 
     source = load_source("multi", s1, s2)
     for s in source: