diff --git a/xmitgcm/demo.py b/xmitgcm/demo.py
new file mode 100644
index 0000000..2c1fb46
--- /dev/null
+++ b/xmitgcm/demo.py
@@ -0,0 +1,20 @@
+
+import os
+
+from .utils import parse_meta_file
+
+
+def collect_meta(path: str):
+
+    output = {}
+    for m in (m for m in os.listdir(path) if m[-5:] == ".meta"):
+        try:
+            meta = parse_meta_file(os.path.join(path, m))
+            if meta["basename"] not in output:
+                output[meta["basename"]] = []
+            meta["filename"] = m
+            output[meta["basename"]].append(meta)
+        except:
+            print(f"Failed parsing: {m}")
+
+    return output
diff --git a/xmitgcm/mds_tilez.py b/xmitgcm/mds_tilez.py
new file mode 100644
index 0000000..c9ce9e7
--- /dev/null
+++ b/xmitgcm/mds_tilez.py
@@ -0,0 +1,250 @@
+
+from collections.abc import Container
+from collections import UserDict
+from dataclasses import dataclass
+import json
+import os.path
+import re
+
+from fsspec.implementations.reference import ReferenceFileSystem
+import numpy as np
+import xarray as xr
+
+from xmitgcm.utils import parse_meta_file
+
+@dataclass()
+class Chunk():
+    """
+    Handles one chunk of one variable. For instance, it could be:
+    S.0000000000.001.001.data
+    S.0000000000.001.001.meta
+
+    metafilename = "/Users/castelao/work/projects/others/MIT_tiles/data/mitgcm/S.0000000000.001.001.meta"
+    """
+    filename: str
+    metadata: dict
+
+    def __fspath__(self):
+        return os.path.join(self.root, self._fdata)
+
+    def __getitem__(self, key):
+        return self._labels[key]
+
+    @property
+    def _labels(self):
+        size = np.prod([(d[2] - d[1] + 1) for d in self.metadata['dimList']])
+        size *= self.metadata['dataprec'].itemsize # f32
+
+        # return {f"{v}/{self.index}": [self.filename, i*size, (i+1)*size] for i,v in enumerate(self.varnames)}
+        return {v: [self.filename, i*size, (i+1)*size] for i,v in enumerate(self.varnames)}
+
+    @property
+    def varnames(self):
+        try:
+            return self._varnames
+        except:
+            if "fldList" in self.metadata:
+                self._varnames = self.metadata['fldList']
+            else:
+                self._varnames = os.path.basename(self.filename).split('.')[0]
+            return self._varnames
+
+    @property
+    def dtype(self):
+        return self.metadata['dataprec'].str
+
+    @property
+    def index(self):
+        idx = []
+        for d in self.metadata['dimList']:
+            idx.append(str((d[1] - 1) //  (d[2] - d[1]+1)))
+
+        return ".".join(idx)
+
+    @property
+    def missing_value(self):
+        if self.metadata['dataprec'].kind == 'i':
+            return int(self.metadata['missingValue'])
+        else:
+            return float(self.metadata['missingValue'])
+
+    @property
+    def shape(self):
+        return tuple(s[0] for s in self.metadata['dimList'])
+
+    @property
+    def chunks(self):
+        return tuple(s[2] for s in self.metadata['dimList'])
+
+    @property
+    def time_step_number(self):
+        return int(self.metadata['timeStepNumber'])
+
+    @staticmethod
+    def from_meta(filename):
+        assert os.path.exists(filename.replace(".meta", ".data"))
+        metadata = parse_meta_file(filename)
+        chunk = Chunk(
+            filename=filename.replace(".meta", ".data"),
+            metadata=metadata)
+        return chunk
+
+
+@dataclass()
+class VarZ():
+    varname: str
+    data: dict
+    chunks: tuple
+    dtype: str
+    fill_value: float
+    shape: tuple
+
+    def __getitem__(self, key):
+        print(f"VarZ.getitem: {key}")
+
+        assert key[:len(self.varname)+1] == f"{self.varname}/"
+        k = key[len(self.varname)+1:]
+        if k == f".zattrs":
+            return self._zattrs()
+        elif k == f".zarray":
+            return self._zarray()
+        else:
+            ti, xyi = k.split(".", 1)
+            ts = sorted(self.data.keys())[int(ti)]
+            return self.data[ts][xyi]
+
+    def __iter__(self):
+        print(f"VarZ.__iter__()")
+        yield from [
+            f"{self.varname}/.zattrs",
+            f"{self.varname}/.zarray",
+        ]
+        idx_t = sorted(self.data.keys())
+        for ts in idx_t:
+            yield from (f"{self.varname}/{idx_t.index(ts)}.{i}"
+                        for i in self.data[ts])
+
+    def _zattrs(self):
+        # How to guess this?
+        if len(self.shape) == 3:
+            dims = ["time", "lon", "lat", "depth"]
+        elif len(self.shape) == 2:
+            dims = ["time", "lon", "lat"]
+
+        return json.dumps({
+            # "_ARRAY_DIMENSIONS": ["lon", "lat", "depth"]
+            "_ARRAY_DIMENSIONS": dims
+        })
+
+    def _zarray(self):
+            return json.dumps({
+                "chunks": [1] + list(self.chunks),
+                "compressor": None, # fixed
+                "dtype": self.dtype,
+                "fill_value": self.fill_value,
+                "filters": None, # fixed
+                "order": "C", # fixed
+                "shape": [len(self.data.keys())] + list(self.shape),
+                "zarr_format": 2 # fixed
+            })
+
+    def push(self, chunk):
+        assert self.varname in chunk.varnames
+        assert self.chunks == chunk.chunks
+        assert self.dtype == chunk.dtype
+        assert self.shape == chunk.shape
+        assert self.fill_value == chunk.missing_value
+
+        if chunk.time_step_number not in self.data:
+            self.data[chunk.time_step_number] = {}
+        self.data[chunk.time_step_number][chunk.index] = chunk[self.varname]
+
+    def push_from_meta(self, filename):
+        d = Chunk.from_meta(filename)
+        self.push(d)
+
+    @staticmethod
+    def from_chunk(chunk, varname=None):
+        if varname is None:
+            assert len(chunk.varnames) == 1
+            varname = chunk.varnames[0]
+
+        assert varname in chunk.varnames
+
+        v = VarZ(varname=varname, data={}, chunks=chunk.chunks,
+                 shape=chunk.shape,
+                 dtype=chunk.dtype, fill_value=chunk.missing_value)
+        v.push(chunk)
+        return v
+
+    @staticmethod
+    def from_meta(filename, varname=None):
+        return VarZ.from_chunk(Chunk.from_meta(filename))
+
+
+class TileZ(UserDict):
+    def __getitem__(self, key):
+        print(f"TileZ.__getitem__(): {key}")
+
+        if key == ".zgroup":
+            return self._zgroup()
+        for v in self.data:
+            if key.startswith(v):
+                return self.data[v][key]
+
+        return super().__getitem__(key)
+
+    def __contains__(self, item):
+        print(f"TileZ.__contains__(): {item}")
+        if item in (".zgroup"):
+            return True
+        for v in self.data:
+            if item in self.data[v]:
+                return True
+        return False
+
+    def __iter__(self):
+        print(f"TileZ.__iter__()")
+        yield from (".zgroup",)
+        for v in self.data:
+            yield from self.data[v].__iter__()
+
+    def _zgroup(self):
+        return json.dumps({"zarr_format": 2})
+
+    def as_dataset(self):
+        fs = ReferenceFileSystem(fo=self, target_protocol='file')
+        mapper = fs.get_mapper("")
+        ds = xr.open_zarr(mapper, consolidated=False)
+        return ds
+
+    def get(self, key, default=None):
+        print(f"TileZ.get(): {key} / {default}")
+        return self.data.get(key, default)
+        #return super().get(*args)
+
+    def len(self):
+        return len([v for v in self])
+
+    def push(self, chunk):
+        for v in chunk.varnames:
+            if v not in self.data:
+                self.data[v] = VarZ.from_chunk(chunk, varname=v)
+            self.data[v].push(chunk)
+
+    def push_from_meta(self, filename):
+        c = Chunk.from_meta(filename)
+        self.push(c)
+
+    def scan(self, path):
+        pattern = re.compile('\w+\.\d+\.meta')
+        filenames = (m for m in os.listdir(path) if pattern.match(m))
+        for mfilename in sorted(filenames):
+            self.push_from_meta(os.path.join(path, mfilename))
+
+    def values(self):
+        print(f"TileZ.values()")
+        yield self._zgroup()
+        for v in self.data:
+            # yield from self.data[v].values()
+            yield from [self.data[v][k] for k in self.data[v]]
diff --git a/xmitgcm/test/test_mds_tilez.py b/xmitgcm/test/test_mds_tilez.py
new file mode 100644
index 0000000..f6857eb
--- /dev/null
+++ b/xmitgcm/test/test_mds_tilez.py
@@ -0,0 +1,36 @@
+
+import json
+import numpy as np
+import pytest
+import tempfile
+
+from xmitgcm.mds_tilez import VarZ
+
+@pytest.fixture(scope="session")
+def var_lat(tmp_path_factory):
+    return tmp_path_factory.mktemp("lat")
+
+
+def test_var_zattrs(var_lat):
+    v = VarZ(var_lat, 'lat')
+
+    attrs = v['.zattrs']
+    attrs = json.loads(attrs)
+    assert "_ARRAY_DIMENSIONS" in attrs
+    assert attrs["_ARRAY_DIMENSIONS"] == ["lat"]
+
+
+def test_var_zarray(var_lat):
+    v = VarZ(var_lat, 'lat')
+
+    attrs = v['.zarray']
+    attrs = json.loads(attrs)
+    assert "chunks" in attrs
+
+def test_var_data(var_lat):
+    v = VarZ(var_lat, 'lat')
+
+    data = v['0']
+    data = np.frombuffer(data)
+    assert np.all(data == np.array([10., 11., 12.]))
+