diff --git a/h5grove/content.py b/h5grove/content.py
index 459d1d6..72c2206 100644
--- a/h5grove/content.py
+++ b/h5grove/content.py
@@ -190,7 +190,7 @@ def data_stats(
         return get_array_stats(data)
 
     def _get_finite_data(self, selection: Selection) -> np.ndarray:
-        data = np.array(self.data(selection), copy=False)  # So it works with scalars
+        data = np.asarray(self.data(selection))  # So it works with scalars
 
         if not np.issubdtype(data.dtype, np.floating):
             return data
@@ -288,6 +288,7 @@ def get_content_from_file(
             fallback=LinkResolution.ONLY_VALID,
         )
     except QueryArgumentError as e:
+        f.close()
         raise create_error(422, str(e))
 
     try:
diff --git a/h5grove/encoders.py b/h5grove/encoders.py
index b9a3316..12a1b44 100644
--- a/h5grove/encoders.py
+++ b/h5grove/encoders.py
@@ -111,7 +111,7 @@ def encode(content: Any, encoding: Optional[str] = "json") -> Response:
             headers={"Content-Type": "application/json"},
         )
 
-    content_array = np.array(content, copy=False)
+    content_array = np.asarray(content)
 
     if encoding == "bin":
         return Response(
@@ -126,21 +126,26 @@ def encode(content: Any, encoding: Optional[str] = "json") -> Response:
             f"Unsupported encoding {encoding} for non-numeric content"
         )
 
-    if encoding == "csv":
+    if encoding == "npy":
         return Response(
-            csv_encode(content_array),
+            npy_encode(content_array),
             headers={
-                "Content-Type": "text/csv",
-                "Content-Disposition": 'attachment; filename="data.csv"',
+                "Content-Type": "application/octet-stream",
+                "Content-Disposition": 'attachment; filename="data.npy"',
             },
         )
 
-    if encoding == "npy":
+    if content_array.ndim == 0:
+        raise QueryArgumentError(
+            f"Unsupported encoding {encoding} for empty and scalar datasets"
+        )
+
+    if encoding == "csv":
         return Response(
-            npy_encode(content_array),
+            csv_encode(content_array),
             headers={
-                "Content-Type": "application/octet-stream",
-                "Content-Disposition": 'attachment; filename="data.npy"',
+                "Content-Type": "text/csv",
+                "Content-Disposition": 'attachment; filename="data.csv"',
             },
         )
 
diff --git a/setup.cfg b/setup.cfg
index 7d4d8d6..14e7687 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -61,7 +61,7 @@ dev =
 	types-contextvars
 	types-dataclasses
 	types-orjson
-	types-pkg-resources
+	types-setuptools
 
 # E501 (line too long) ignored for now
 # E203 and W503 incompatible with black formatting (https://black.readthedocs.io/en/stable/compatible_configs.html#flake8)
diff --git a/test/base_test.py b/test/base_test.py
index 525006f..154d161 100644
--- a/test/base_test.py
+++ b/test/base_test.py
@@ -77,6 +77,26 @@ def test_data_on_array_with_format(self, server, format_arg):
 
         assert np.array_equal(retrieved_data, data)
 
+    @pytest.mark.parametrize("format_arg", ("json", "bin", "npy"))
+    def test_data_on_scalar_with_format(self, server, format_arg):
+        """Test /data/ endpoint on scalar dataset"""
+        # Test condition
+        tested_h5entity_path = "/entry/scalar"
+        data = 5
+
+        filename = "test.h5"
+        with h5py.File(server.served_directory / filename, mode="w") as h5file:
+            dset = h5file.create_dataset(tested_h5entity_path, data=data)
+            dtype = dset.dtype
+            shape = dset.shape
+
+        response = server.get(
+            f"/data/?{urlencode({'file': filename, 'path': tested_h5entity_path, 'format': format_arg})}"
+        )
+        retrieved_data = decode_array_response(response, format_arg, dtype.str, shape)
+
+        assert np.array_equal(retrieved_data, data)
+
     @pytest.mark.parametrize("format_arg", ("npy", "bin"))
     def test_data_on_array_with_dtype_safe(
         self,
@@ -114,7 +134,7 @@ def test_data_on_slice_with_format_and_flatten(self, server, format_arg):
         response = server.get(
             f"/data/?{urlencode({'file': filename, 'path': tested_h5entity_path, 'selection': '100,0', 'format': format_arg, 'flatten': True})}"
         )
-        retrieved_data = np.array(decode_response(response, format_arg))
+        retrieved_data = np.asarray(decode_response(response, format_arg))
 
         assert retrieved_data - data[100, 0] < 1e-8
 
@@ -575,3 +595,20 @@ def test_422_on_invalid_query_arg(self, server):
             f"/meta/?file={filename}&path={path}&resolve_links={invalid_link_resolution}",
             422,
         )
+
+    @pytest.mark.parametrize("format_arg", ("csv", "tiff"))
+    def test_422_on_format_incompatible_with_empty_or_scalar_datasets(
+        self, server, format_arg
+    ):
+        filename = "test.h5"
+
+        with h5py.File(server.served_directory / filename, mode="w") as h5file:
+            h5file["scalar"] = 55
+            h5file["empty"] = h5py.Empty(dtype="<4f")
+
+        server.assert_error_code(
+            f"/data/?file={filename}&path=/scalar&format={format_arg}", 422
+        )
+        server.assert_error_code(
+            f"/data/?file={filename}&path=/empty&format={format_arg}", 422
+        )
diff --git a/test/utils.py b/test/utils.py
index 301941e..fcedcf2 100644
--- a/test/utils.py
+++ b/test/utils.py
@@ -64,7 +64,7 @@ def decode_array_response(
         assert content_type == "application/octet-stream"
         return np.frombuffer(response.content, dtype=dtype).reshape(shape)
 
-    return np.array(decode_response(response, format), copy=False)
+    return np.asarray(decode_response(response, format))
 
 
 def assert_error_response(response: Response, error_code: int):