diff --git a/third_party/endpoints/message_parser.py b/third_party/endpoints/message_parser.py
new file mode 100644
index 0000000..28d6f47
--- /dev/null
+++ b/third_party/endpoints/message_parser.py
@@ -0,0 +1,227 @@
+# Copyright 2016 Google Inc. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Describe ProtoRPC Messages in JSON Schema.
+
+Add protorpc.message subclasses to MessageTypeToJsonSchema and get a JSON
+Schema description of all the messages.
+"""
+
+# pylint: disable=g-bad-name
+from __future__ import absolute_import
+
+import re
+
+from . import message_types
+from . import messages
+
+__all__ = ['MessageTypeToJsonSchema']
+
+
+class MessageTypeToJsonSchema(object):
+  """Describe ProtoRPC messages in JSON Schema.
+
+  Add protorpc.message subclasses to MessageTypeToJsonSchema and get a JSON
+  Schema description of all the messages. MessageTypeToJsonSchema handles
+  all the types of fields that can appear in a message.
+  """
+
+  # Field to schema type and format.  If the field maps to tuple, the
+  # first entry is set as the type, the second the format (or left alone if
+  # None).  If the field maps to a dictionary, we'll grab the value from the
+  # field's Variant in that dictionary.
+  # The variant dictionary should include an element that None maps to,
+  # to fall back on as a default.
+  __FIELD_TO_SCHEMA_TYPE_MAP = {
+      messages.IntegerField: {messages.Variant.INT32: ('integer', 'int32'),
+                              messages.Variant.INT64: ('string', 'int64'),
+                              messages.Variant.UINT32: ('integer', 'uint32'),
+                              messages.Variant.UINT64: ('string', 'uint64'),
+                              messages.Variant.SINT32: ('integer', 'int32'),
+                              messages.Variant.SINT64: ('string', 'int64'),
+                              None: ('integer', 'int64')},
+      messages.FloatField: {messages.Variant.FLOAT: ('number', 'float'),
+                            messages.Variant.DOUBLE: ('number', 'double'),
+                            None: ('number', 'float')},
+      messages.BooleanField: ('boolean', None),
+      messages.BytesField: ('string', 'byte'),
+      message_types.DateTimeField: ('string', 'date-time'),
+      messages.StringField: ('string', None),
+      messages.MessageField: ('object', None),
+      messages.EnumField: ('string', None),
+  }
+
+  __DEFAULT_SCHEMA_TYPE = ('string', None)
+
+  def __init__(self):
+    # A map of schema ids to schemas.
+    self.__schemas = {}
+
+    # A map from schema id to non-normalized definition name.
+    self.__normalized_names = {}
+
+  def add_message(self, message_type):
+    """Add a new message.
+
+    Args:
+      message_type: protorpc.message.Message class to be parsed.
+
+    Returns:
+      string, The JSON Schema id.
+
+    Raises:
+      KeyError if the Schema id for this message_type would collide with the
+      Schema id of a different message_type that was already added.
+    """
+    name = self.__normalized_name(message_type)
+    if name not in self.__schemas:
+      # Set a placeholder to prevent infinite recursion.
+      self.__schemas[name] = None
+      schema = self.__message_to_schema(message_type)
+      self.__schemas[name] = schema
+    return name
+
+  def ref_for_message_type(self, message_type):
+    """Returns the JSON Schema id for the given message.
+
+    Args:
+      message_type: protorpc.message.Message class to be parsed.
+
+    Returns:
+      string, The JSON Schema id.
+
+    Raises:
+      KeyError: if the message hasn't been parsed via add_message().
+    """
+    name = self.__normalized_name(message_type)
+    if name not in self.__schemas:
+      raise KeyError('Message has not been parsed: %s', name)
+    return name
+
+  def schemas(self):
+    """Returns the JSON Schema of all the messages.
+
+    Returns:
+      object: JSON Schema description of all messages.
+    """
+    return self.__schemas.copy()
+
+  def __normalized_name(self, message_type):
+    """Normalized schema name.
+
+    Generate a normalized schema name, taking the class name and stripping out
+    everything but alphanumerics, and camel casing the remaining words.
+    A normalized schema name is a name that matches [a-zA-Z][a-zA-Z0-9]*
+
+    Args:
+      message_type: protorpc.message.Message class being parsed.
+
+    Returns:
+      A string, the normalized schema name.
+
+    Raises:
+      KeyError: A collision was found between normalized names.
+    """
+    # Normalization is applied to match the constraints that Discovery applies
+    # to Schema names.
+    name = message_type.definition_name()
+
+    split_name = re.split(r'[^0-9a-zA-Z]', name)
+    normalized = ''.join(
+        part[0].upper() + part[1:] for part in split_name if part)
+
+    previous = self.__normalized_names.get(normalized)
+    if previous:
+      if previous != name:
+        raise KeyError('Both %s and %s normalize to the same schema name: %s' %
+                       (name, previous, normalized))
+    else:
+      self.__normalized_names[normalized] = name
+
+    return normalized
+
+  def __message_to_schema(self, message_type):
+    """Parse a single message into JSON Schema.
+
+    Will recursively descend the message structure
+    and also parse other messages references via MessageFields.
+
+    Args:
+      message_type: protorpc.messages.Message class to parse.
+
+    Returns:
+      An object representation of the schema.
+    """
+    name = self.__normalized_name(message_type)
+    schema = {
+        'id': name,
+        'type': 'object',
+        }
+    if message_type.__doc__:
+      schema['description'] = message_type.__doc__
+    properties = {}
+    for field in message_type.all_fields():
+      descriptor = {}
+      # Info about the type of this field.  This is either merged with
+      # the descriptor or it's placed within the descriptor's 'items'
+      # property, depending on whether this is a repeated field or not.
+      type_info = {}
+
+      if type(field) == messages.MessageField:
+        field_type = field.type().__class__
+        type_info['$ref'] = self.add_message(field_type)
+        if field_type.__doc__:
+          descriptor['description'] = field_type.__doc__
+      else:
+        schema_type = self.__FIELD_TO_SCHEMA_TYPE_MAP.get(
+            type(field), self.__DEFAULT_SCHEMA_TYPE)
+        # If the map pointed to a dictionary, check if the field's variant
+        # is in that dictionary and use the type specified there.
+        if isinstance(schema_type, dict):
+          variant_map = schema_type
+          variant = getattr(field, 'variant', None)
+          if variant in variant_map:
+            schema_type = variant_map[variant]
+          else:
+            # The variant map needs to specify a default value, mapped by None.
+            schema_type = variant_map[None]
+        type_info['type'] = schema_type[0]
+        if schema_type[1]:
+          type_info['format'] = schema_type[1]
+
+      if type(field) == messages.EnumField:
+        sorted_enums = sorted([enum_info for enum_info in field.type],
+                              key=lambda enum_info: enum_info.number)
+        type_info['enum'] = [enum_info.name for enum_info in sorted_enums]
+
+      if field.required:
+        descriptor['required'] = True
+
+      if field.default:
+        if type(field) == messages.EnumField:
+          descriptor['default'] = str(field.default)
+        else:
+          descriptor['default'] = field.default
+
+      if field.repeated:
+        descriptor['items'] = type_info
+        descriptor['type'] = 'array'
+      else:
+        descriptor.update(type_info)
+
+      properties[field.name] = descriptor
+
+    schema['properties'] = properties
+
+    return schema