diff --git a/tools/python/mwm/dump_mwm.py b/tools/python/mwm/dump_mwm.py
new file mode 100755
index 0000000000..bc86ba53cd
--- /dev/null
+++ b/tools/python/mwm/dump_mwm.py
@@ -0,0 +1,24 @@
+#!/usr/bin/python
+import sys, os.path
+import itertools
+from mwm import MWM
+
+if len(sys.argv) < 2:
+  print 'Dumps some MWM structures.'
+  print 'Usage: {0} <country.mwm>'.format(sys.argv[0])
+  sys.exit(1)
+
+mwm = MWM(open(sys.argv[1], 'rb'))
+mwm.read_types(os.path.join(os.path.dirname(sys.argv[0]), '..', '..', '..', 'data', 'types.txt'))
+print 'Tags:'
+for tag, value in mwm.tags.iteritems():
+  print '  {0:<8}: offs {1:9} len {2:8}'.format(tag, value[0], value[1])
+print 'Version:', mwm.read_version()
+print 'Header:', mwm.read_header()
+print 'Metadata count:', len(mwm.read_metadata())
+cross = mwm.read_crossmwm()
+if cross:
+  print 'Outgoing points:', len(cross['out']), 'incoming:', len(cross['in'])
+  print 'Outgoing regions:', set(cross['neighbours'])
+for feature in itertools.islice(mwm.iter_features(), 10):
+  print feature
diff --git a/tools/python/mwm/mwm.py b/tools/python/mwm/mwm.py
new file mode 100644
index 0000000000..1ad6851331
--- /dev/null
+++ b/tools/python/mwm/mwm.py
@@ -0,0 +1,382 @@
+# MWM Reader Module
+import struct
+import math
+
+# Unprocessed sections: geomN, trgN, idx, sdx (search index), addr (search address), offs (feature offsets), dat (!)
+# Routing sections: mercedes (matrix), daewoo (edge data), infinity (edge id), skoda (shortcuts), chrysler (cross context), ftseg, node2ftseg
+# (these mostly are succinct structures, except chrysler and node2ftseg, so no use trying to load them here)
+
+class MWM:
+  languages = ["default",
+    "en", "ja", "fr", "ko_rm", "ar", "de", "int_name", "ru", "sv", "zh", "fi", "be", "ka", "ko",
+    "he", "nl", "ga", "ja_rm", "el", "it", "es", "zh_pinyin", "th", "cy", "sr", "uk", "ca", "hu",
+    "hsb", "eu", "fa", "br", "pl", "hy", "kn", "sl", "ro", "sq", "am", "fy", "cs", "gd", "sk",
+    "af", "ja_kana", "lb", "pt", "hr", "fur", "vi", "tr", "bg", "eo", "lt", "la", "kk", "gsw",
+    "et", "ku", "mn", "mk", "lv", "hi"]
+
+  metadata = ["0",
+    "cuisine", "open_hours", "phone_number", "fax_number", "stars",
+    "operator", "url", "website", "internet", "ele",
+    "turn_lanes", "turn_lanes_forward", "turn_lanes_backward", "email", "postcode",
+    "wikipedia", "maxspeed", "flats", "height", "min_height",
+    "denomination", "building_levels"
+  ]
+
+  def __init__(self, f):
+    self.f = f
+    self.coord_size = None
+    self.base_point = (0, 0)
+    self.read_info()
+    self.type_mapping = []
+
+  def read_types(self, filename):
+    with open(filename, 'r') as ft:
+      for line in ft:
+        if len(line.strip()) > 0:
+          self.type_mapping.append(line.strip())
+
+  def read_info(self):
+    self.f.seek(0)
+    self.f.seek(self.read_uint(8))
+    cnt = self.read_varuint()
+    self.tags = {}
+    for i in range(cnt):
+      name = self.read_string(True)
+      offset = self.read_varuint()
+      length = self.read_varuint()
+      self.tags[name] = (offset, length)
+
+  def has_tag(self, tag):
+    return tag in self.tags and self.tags[tag][1] > 0
+
+  def seek_tag(self, tag):
+    self.f.seek(self.tags[tag][0])
+
+  def inside_tag(self, tag):
+    pos = self.tag_position(tag)
+    return pos >= 0 and pos < self.tags[tag][1]
+
+  def tag_position(self, tag):
+    return self.f.tell() - self.tags[tag][0]
+
+  def read_version(self):
+    """Reads 'version' section."""
+    self.seek_tag('version')
+    self.f.read(4) # skip prolog
+    fmt = self.read_varuint() + 1
+    version = self.read_varuint()
+    return { 'fmt': fmt, 'version': version }
+
+  def read_header(self):
+    """Reads 'header' section."""
+    if not self.has_tag('header'):
+      # Stub for routing files
+      self.coord_size = (1 << 30) - 1
+      return {}
+    self.seek_tag('header')
+    result = {}
+    coord_bits = self.read_varuint()
+    self.coord_size = (1 << coord_bits) - 1
+    self.base_point = self.read_coord(convert=False)
+    result['basePoint'] = self.to_4326(self.base_point)
+    result['bounds'] = self.read_bounds()
+    result['scales'] = self.read_uint_array()
+    langs = self.read_uint_array()
+    for i in range(len(langs)):
+      if i < len(self.languages):
+        langs[i] = self.languages[langs[i]]
+    result['langs'] = langs
+    map_type = self.read_varint()
+    if map_type == 0:
+      result['mapType'] = 'world'
+    elif map_type == 1:
+      result['mapType'] = 'worldcoasts'
+    elif map_type == 2:
+      result['mapType'] = 'country'
+    else:
+      result['mapType'] = 'unknown: {0}'.format(map_type)
+    return result
+
+  # COMPLEX READERS
+
+  def read_metadata(self):
+    """Reads 'meta' and 'metaidx' sections."""
+    if not self.has_tag('metaidx'):
+      return {}
+    # Metadata format is different since v8
+    fmt = self.read_version()['fmt']
+    # First, read metaidx, to match featureId <-> metadata
+    self.seek_tag('metaidx')
+    ftid_meta = []
+    while self.inside_tag('metaidx'):
+      ftid = self.read_uint(4)
+      moffs = self.read_uint(4)
+      ftid_meta.append((moffs, ftid))
+    # Sort ftid_meta array
+    ftid_meta.sort(key=lambda x: x[0])
+    ftpos = 0
+    # Now read metadata
+    self.seek_tag('meta')
+    metadatar = {}
+    while self.inside_tag('meta'):
+      tag_pos = self.tag_position('meta')
+      fields = {}
+      if fmt >= 8:
+        sz = self.read_varuint()
+        if sz:
+          for i in range(sz):
+            t = self.read_varuint()
+            t = self.metadata[t] if t < len(self.metadata) else str(t)
+            fields[t] = self.read_string()
+      else:
+        while True:
+          t = self.read_uint(1)
+          is_last = t & 0x80 > 0
+          t = t & 0x7f
+          t = self.metadata[t] if t < len(self.metadata) else str(t)
+          l = self.read_uint(1)
+          fields[t] = self.f.read(l)
+          if is_last:
+            break
+
+      if len(fields):
+        while ftpos < len(ftid_meta) and ftid_meta[ftpos][0] < tag_pos:
+          ftpos += 1
+        if ftpos < len(ftid_meta):
+          if ftid_meta[ftpos][0] == tag_pos:
+            metadatar[ftid_meta[ftpos][1]] = fields
+    return metadatar
+
+  def read_crossmwm(self):
+    """Reads 'chrysler' section (cross-mwm routing table)."""
+    if not self.has_tag('chrysler'):
+      return {}
+    self.seek_tag('chrysler')
+    # Ingoing nodes: array of (nodeId, coord) tuples
+    incomingCount = self.read_uint(4)
+    incoming = []
+    for i in range(incomingCount):
+      nodeId = self.read_uint(4)
+      point = self.read_coord(False)
+      incoming.append((nodeId, point))
+    # Outgoing nodes: array of (nodeId, coord, outIndex) tuples
+    # outIndex is an index in neighbours array
+    outgoingCount = self.read_uint(4)
+    outgoing = []
+    for i in range(outgoingCount):
+      nodeId = self.read_uint(4)
+      point = self.read_coord(False)
+      outIndex = self.read_uint(1)
+      outgoing.append((nodeId, point, outIndex))
+    # Adjacency matrix: costs of routes for each (incoming, outgoing) tuple
+    matrix = []
+    for i in range(incomingCount):
+      sub = []
+      for j in range(outgoingCount):
+        sub.append(self.read_uint(4))
+      matrix.append(sub)
+    # List of mwms to which leads each outgoing node
+    neighboursCount = self.read_uint(4)
+    neighbours = []
+    for i in range(neighboursCount):
+      size = self.read_uint(4)
+      neighbours.append(self.f.read(size))
+    return { 'in': incoming, 'out': outgoing, 'matrix': matrix, 'neighbours': neighbours }
+
+  class GeomType:
+    POINT = 0
+    LINE = 1 << 5
+    AREA = 1 << 6
+    POINT_EX = 3 << 5
+
+  class OsmIdCode:
+    NODE = 0x4000000000000000;
+    WAY = 0x8000000000000000;
+    RELATION = 0xC000000000000000;
+    RESET = ~(NODE | WAY | RELATION);
+
+  def iter_features(self):
+    """Reads 'dat' section."""
+    if not self.has_tag('dat'):
+      return
+    # TODO: read 'offs'?
+    self.seek_tag('dat')
+    while self.inside_tag('dat'):
+      feature = {}
+      feature_size = self.read_varuint()
+      next_feature = self.f.tell() + feature_size
+      feature['size'] = feature_size
+
+      # Header
+      header = {}
+      header_bits = self.read_uint(1)
+      types_count = (header_bits & 0x07) + 1
+      has_name = header_bits & 0x08 > 0
+      has_layer = header_bits & 0x10 > 0
+      has_addinfo = header_bits & 0x80 > 0
+      geom_type = header_bits & 0x60
+      types = []
+      for i in range(types_count):
+        type_id = self.read_varuint()
+        if type_id < len(self.type_mapping):
+          types.append(self.type_mapping[type_id])
+        else:
+          types.append(str(type_id))
+      header['types'] = types
+      if has_name:
+        header['name'] = self.read_multilang()
+      if has_layer:
+        header['layer'] = self.read_uint(1)
+      if has_addinfo:
+        if geom_type == MWM.GeomType.POINT:
+          header['rank'] = self.read_uint(1)
+        elif geom_type == MWM.GeomType.LINE:
+          header['ref'] = self.read_string()
+        elif geom_type == MWM.GeomType.AREA or geom_type == MWM.GeomType.POINT_EX:
+          header['house'] = self.read_numeric_string()
+      feature['header'] = header
+
+      # Geometry
+      geometry = {}
+      if geom_type == MWM.GeomType.POINT or geom_type == MWM.GeomType.POINT_EX:
+        geometry['type'] = 'Point'
+      elif geom_type == MWM.GeomType.LINE:
+        geometry['type'] = 'LineString'
+      elif geom_type == MWM.GeomType.AREA:
+        geometry['type'] = 'Polygon'
+      if geom_type == MWM.GeomType.POINT:
+        geometry['coordinates'] = list(self.read_coord())
+
+      # (flipping table emoticon)
+      feature['geometry'] = geometry
+      if False:
+        if geom_type != MWM.GeomType.POINT:
+          polygon_count = self.read_varuint()
+          polygons = []
+          for i in range(polygon_count):
+            count = self.read_varuint()
+            buf = self.f.read(count)
+            # TODO: decode
+          geometry['coordinates'] = polygons
+          feature['coastCell'] = self.read_varint()
+        
+        # OSM IDs
+        count = self.read_varuint()
+        osmids = []
+        for i in range(count):
+          encid = self.read_uint(8)
+          if encid & MWM.OsmIdCode.NODE == MWM.OsmIdCode.NODE:
+            typ = 'n'
+          elif encid & MWM.OsmIdCode.WAY == MWM.OsmIdCode.WAY:
+            typ = 'w'
+          elif encid & MWM.OsmIdCode.RELATION == MWM.OsmIdCode.RELATION:
+            typ = 'r'
+          else:
+            typ = ''
+          osmids.append('{0}{1}'.format(typ, encid & MWM.OsmIdCode.RESET))
+        feature['osmIds'] = osmids
+
+      if self.f.tell() > next_feature:
+        raise Exception('Feature parsing error, read too much')
+      yield feature
+      self.f.seek(next_feature)
+
+  # BITWISE READERS
+
+  def read_uint(self, bytelen=1):
+    if bytelen == 1:
+      fmt = 'B'
+    elif bytelen == 2:
+      fmt = 'H'
+    elif bytelen == 4:
+      fmt = 'I'
+    elif bytelen == 8:
+      fmt = 'Q'
+    else:
+      raise Exception('Bytelen {0} is not supported'.format(bytelen))
+    res = struct.unpack(fmt, self.f.read(bytelen))
+    return res[0]
+
+  def read_varuint(self):
+    res = 0
+    shift = 0
+    more = True
+    while more:
+      b = self.f.read(1)
+      if not b:
+        return res
+      res |= (ord(b[0]) & 0x7F) << shift
+      shift += 7
+      more = ord(b[0]) >= 0x80
+    return res
+
+  def read_varint(self):
+    uint = self.read_varuint()
+    res = uint >> 1
+    return res if uint & 1 == 0 else -res
+
+  def mwm_unshuffle(self, x):
+    x = ((x & 0x22222222) << 1) | ((x >> 1) & 0x22222222) | (x & 0x99999999)
+    x = ((x & 0x0C0C0C0C) << 2) | ((x >> 2) & 0x0C0C0C0C) | (x & 0xC3C3C3C3)
+    x = ((x & 0x00F000F0) << 4) | ((x >> 4) & 0x00F000F0) | (x & 0xF00FF00F)
+    x = ((x & 0x0000FF00) << 8) | ((x >> 8) & 0x0000FF00) | (x & 0xFF0000FF)
+    return x
+
+  def mwm_bitwise_split(self, v):
+    hi = self.mwm_unshuffle(v >> 32)
+    lo = self.mwm_unshuffle(v & 0xFFFFFFFF)
+    x = ((hi & 0xFFFF) << 16) | (lo & 0xFFFF);
+    y =     (hi & 0xFFFF0000) | (lo >> 16);
+    return (x, y)
+
+  def read_point(self, packed=True):
+    """Reads an unsigned point, returns (x, y)."""
+    if packed:
+      u = self.read_varuint()
+    else:
+      u = self.read_uint(8)
+    return self.mwm_bitwise_split(u)
+
+  def to_4326(self, point):
+    if self.coord_size is None:
+      raise Exception('Call read_header() first.')
+    merc_bounds = (-180, -180, 180, 180) # Xmin, Ymin, Xmax, Ymax
+    x = point[0] * (merc_bounds[2] - merc_bounds[0]) / self.coord_size + merc_bounds[0]
+    y = point[1] * (merc_bounds[3] - merc_bounds[1]) / self.coord_size + merc_bounds[1]
+    y = 360.0 * math.atan(math.tanh(y * math.pi / 360.0)) / math.pi
+    return (x, y)
+
+  def read_coord(self, packed=True, convert=True):
+    """Reads a pair of coords in degrees mercator, returns (lon, lat)."""
+    upoint = self.read_point(packed)
+    point = (upoint[0] + self.base_point[0], upoint[1] + self.base_point[1])
+    return self.to_4326(point) if convert else point
+
+  def read_bounds(self):
+    """Reads mercator bounds, returns (min_lon, min_lat, max_lon, max_lat)."""
+    rmin = self.read_coord()
+    rmax = self.read_coord()
+    return (rmin[0], rmin[1], rmax[0], rmax[1])
+
+  def read_string(self, plain=False):
+    length = self.read_varuint() + (0 if plain else 1)
+    return self.f.read(length)
+
+  def read_uint_array(self):
+    length = self.read_varuint()
+    result = []
+    for i in range(length):
+      result.append(self.read_varuint())
+    return result
+
+  def read_numeric_string(self):
+    sz = self.read_varuint()
+    if sz & 1 != 0:
+      return str(sz >> 1)
+    sz = (sz >> 1) + 1
+    return self.f.read(sz)
+
+  def read_multilang(self):
+    s = self.read_string()
+    # TODO!
+    return s