mypy-protobuf: Generate mypy stub files from protobuf specs

CI pypi license

2.10 is the last version of mypy-protobuf which supports targeting python 2.7.

Built originally with love at Dropbox

See Changelog for recent changes.

Requirements to run mypy-protobuf

Earlier releases might work, but aren't tested

Requirements to run typecheckers on stubs generated by mypy-protobuf

Earlier releases might work, but aren't tested

To run typecheckers on code generated with grpc plugin - you'll additionally need

Earlier releases might work, but aren't tested

Other configurations may work, but are not continuously tested currently. We would be open to expanding this list - file an issue on the issue tracker.

Installation

The plugin can be installed with

pip3 install mypy-protobuf

To install unreleased

REV=main  # or whichever unreleased git rev you'd like
pip3 install git+https://github.com/nipunn1313/mypy-protobuf.git@$REV

# For older (1.x) versions of mypy protobuf - you may need
pip3 install git+https://github.com/nipunn1313/mypy-protobuf.git@$REV#subdirectory=python

In order to run mypy on the generated code, you'll need to install

pip3 install mypy>=0.910 types-protobuf>=0.1.14

Usage

On posix, protoc-gen-mypy is installed to python's executable bin. Assuming that's on your $PATH, you can run

protoc --python_out=output/location --mypy_out=output/location

Alternately, you can explicitly provide the path:

protoc --plugin=protoc-gen-mypy=path/to/protoc-gen-mypy --python_out=output/location --mypy_out=output/location

Check the version number with

> protoc-gen-mypy --version

Implementation

The implementation of the plugin is in mypy_protobuf/main.py, which installs to an executable protoc-gen-mypy. On windows it installs to protoc-gen-mypy.exe

Features

See Changelog for full listing

Differences between the protobuf compiler (--pyi_out) and mypy-protobuf

(As of 11/17/2025)

Examples

mypy-protobuf:

"""
@generated by mypy-protobuf.  Do not edit manually!
isort:skip_file
Edition version of proto2 file"""

import builtins
import google.protobuf.descriptor
import google.protobuf.message
import sys
import typing

if sys.version_info >= (3, 10):
    import typing as typing_extensions
else:
    import typing_extensions

DESCRIPTOR: google.protobuf.descriptor.FileDescriptor

@typing.final
class Editions2024SubMessage(google.protobuf.message.Message):
    DESCRIPTOR: google.protobuf.descriptor.Descriptor

    THING_FIELD_NUMBER: builtins.int
    thing: builtins.str
    def __init__(
        self,
        *,
        thing: builtins.str | None = ...,
    ) -> None: ...
    _HasFieldArgType: typing_extensions.TypeAlias = typing.Literal["thing", b"thing"]
    def HasField(self, field_name: _HasFieldArgType) -> builtins.bool: ...
    _ClearFieldArgType: typing_extensions.TypeAlias = typing.Literal["thing", b"thing"]
    def ClearField(self, field_name: _ClearFieldArgType) -> None: ...

Global___Editions2024SubMessage: typing_extensions.TypeAlias = Editions2024SubMessage

@typing.final
class Editions2024Test(google.protobuf.message.Message):
    DESCRIPTOR: google.protobuf.descriptor.Descriptor

    LEGACY_FIELD_NUMBER: builtins.int
    EXPLICIT_SINGULAR_FIELD_NUMBER: builtins.int
    MESSAGE_FIELD_FIELD_NUMBER: builtins.int
    IMPLICIT_SINGULAR_FIELD_NUMBER: builtins.int
    DEFAULT_SINGULAR_FIELD_NUMBER: builtins.int
    legacy: builtins.str
    """Expect to be always set"""
    explicit_singular: builtins.str
    """Expect HasField generated"""
    implicit_singular: builtins.str
    """Expect implicit field presence, no HasField generated"""
    default_singular: builtins.str
    """Not set, should default to EXPLICIT"""
    @property
    def message_field(self) -> Global___Editions2024SubMessage:
        """Expect HasField generated?"""

    def __init__(
        self,
        *,
        legacy: builtins.str | None = ...,
        explicit_singular: builtins.str | None = ...,
        message_field: Global___Editions2024SubMessage | None = ...,
        implicit_singular: builtins.str = ...,
        default_singular: builtins.str | None = ...,
    ) -> None: ...
    _HasFieldArgType: typing_extensions.TypeAlias = typing.Literal["default_singular", b"default_singular", "explicit_singular", b"explicit_singular", "legacy", b"legacy", "message_field", b"message_field"]
    def HasField(self, field_name: _HasFieldArgType) -> builtins.bool: ...
    _ClearFieldArgType: typing_extensions.TypeAlias = typing.Literal["default_singular", b"default_singular", "explicit_singular", b"explicit_singular", "implicit_singular", b"implicit_singular", "legacy", b"legacy", "message_field", b"message_field"]
    def ClearField(self, field_name: _ClearFieldArgType) -> None: ...

Global___Editions2024Test: typing_extensions.TypeAlias = Editions2024Test

Builtin pyi generator:

from google.protobuf import descriptor as _descriptor
from google.protobuf import message as _message
from collections.abc import Mapping as _Mapping
from typing import ClassVar as _ClassVar, Optional as _Optional, Union as _Union

DESCRIPTOR: _descriptor.FileDescriptor

class Editions2024SubMessage(_message.Message):
    __slots__ = ()
    THING_FIELD_NUMBER: _ClassVar[int]
    thing: str
    def __init__(self, thing: _Optional[str] = ...) -> None: ...

class Editions2024Test(_message.Message):
    __slots__ = ()
    LEGACY_FIELD_NUMBER: _ClassVar[int]
    EXPLICIT_SINGULAR_FIELD_NUMBER: _ClassVar[int]
    MESSAGE_FIELD_FIELD_NUMBER: _ClassVar[int]
    IMPLICIT_SINGULAR_FIELD_NUMBER: _ClassVar[int]
    DEFAULT_SINGULAR_FIELD_NUMBER: _ClassVar[int]
    legacy: str
    explicit_singular: str
    message_field: Editions2024SubMessage
    implicit_singular: str
    default_singular: str
    def __init__(self, legacy: _Optional[str] = ..., explicit_singular: _Optional[str] = ..., message_field: _Optional[_Union[Editions2024SubMessage, _Mapping]] = ..., implicit_singular: _Optional[str] = ..., default_singular: _Optional[str] = ...) -> None: ...

Bring comments from .proto files to docstrings in .pyi files

Comments in the .proto files on messages, fields, enums, enum variants, extensions, services, and methods will appear as docstrings in .pyi files. Useful in IDEs for showing completions with comments.

Types enum int values more strongly

Enum int values produce stubs which wrap the int values in NewType

enum MyEnum {
  HELLO = 0;
  WORLD = 1;
}

Will yield an enum type wrapper whose methods type to MyEnum.ValueType (a NewType(int) rather than int. This allows mypy to catch bugs where the wrong enum value is being used.

Calling code may be typed as follows.

In python >= 3.7

# May need [PEP 563](https://www.python.org/dev/peps/pep-0563/) to postpone evaluation of annotations
# from __future__ import annotations  # Not needed with python>=3.11 or protobuf>=3.20.0
def f(x: MyEnum.ValueType):
    print(x)
f(MyEnum.Value("HELLO"))

With protobuf <= 3.20.0, for usages of cast, the type of x must be quoted After protobuf >= 3.20.0 - ValueType exists in the python code and quotes aren't needed until upstream protobuf includes ValueType

cast('MyEnum.ValueType', x)

Similarly, for type aliases with protobuf < 3.20.0, you must either quote the type or hide it behind TYPE_CHECKING

from typing import Tuple, TYPE_CHECKING
HELLO = Tuple['MyEnum.ValueType', 'MyEnum.ValueType']
if TYPE_CHECKING:
    HELLO = Tuple[MyEnum.ValueType, MyEnum.ValueType]

Enum int impl details

mypy-protobuf autogenerates an instance of the EnumTypeWrapper as follows.

class _MyEnum:
    ValueType = typing.NewType('ValueType', builtins.int)
    V: typing_extensions.TypeAlias = ValueType
class _MyEnumEnumTypeWrapper(google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[_MyEnum.ValueType], builtins.type):
    DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
    HELLO: _MyEnum.ValueType  # 0
    WORLD: _MyEnum.ValueType  # 1
class MyEnum(_MyEnum, metaclass=_MyEnumEnumTypeWrapper):
    pass

HELLO: MyEnum.ValueType  # 0
WORLD: MyEnum.ValueType  # 1

_MyEnumEnumTypeWrapper extends the EnumTypeWrapper to take/return MyEnum.ValueType rather than int MyEnum is an instance of the EnumTypeWrapper.

Supports generating type wrappers for fields and maps

M.proto

message M {
  uint32 user_id = 1 [(mypy_protobuf.options).casttype="mymod.UserId"];
  map<uint32, string> email_by_uid = 2 [
    (mypy_protobuf.options).keytype="path/to/mymod.UserId",
    (mypy_protobuf.options).valuetype="path/to/mymod.Email"
  ];
}

mymod.py

UserId = NewType("UserId", int)
Email = NewType("Email", Text)

py_generic_services

If py_generic_services is set in your proto file, then mypy-protobuf will generate service stubs. If you want GRPC stubs instead - use the GRPC instructions.

readable_stubs

If readable_stubs is set, mypy-protobuf will generate easier-to-read stubs. The downside to this approach - is that it's possible to generate stubs which do not pass mypy - particularly in the case of name collisions. mypy-protobuf defaults to generating stubs with fully qualified imports and mangled global-level identifiers to defend against name collisions between global identifiers and field names.

If you're ok with this risk, try it out!

protoc --python_out=output/location --mypy_out=readable_stubs:output/location

relax_strict_optional_primitives

If you are using proto3, then primitives cannot be represented as NULL on the wire - only as their zero value. By default mypy-protobuf types message constructors to have non-nullable primitives (eg int instead of Optional[int]). python-protobuf itself will internally convert None -> zero value. If you intentionally want to use this behavior, set this flag! We recommend avoiding this, as it can lead to developer error - confusing NULL and 0 as distinct on the wire. However, it may be helpful when migrating existing proto2 code, where the distinction is meaningful

protoc --python_out=output/location --mypy_out=relax_strict_optional_primitives:output/location

use_default_deprecation_warnings

By default mypy-protobuf will pull the leading and trailing comments from the deprecation option definition, and insert it into the deprecation warning. This option will instead use a standard deprecation warning instead of comments.

protoc --python_out=output/location --mypy_out=use_default_deprecation_warning:output/location

generate_concrete_servicer_stubs

By default mypy-protobuf will output servicer stubs with abstract methods. To output concrete stubs, set this option

protoc --python_out=output/location --mypy_grpc_out=generate_concrete_servicer_stubs:output/location

sync_only/async_only

By default, generated GRPC stubs are compatible with both sync and async variants. If you only want sync or async GRPC stubs, use this option:

protoc --python_out=output/location --mypy_grpc_out=sync_only:output/location

or

protoc --python_out=output/location --mypy_grpc_out=async_only:output/location

Output suppression

To suppress output, you can run

protoc --python_out=output/location --mypy_out=quiet:output/location

GRPC

This plugin provides stubs generation for grpcio generated code.

protoc \
    --python_out=output/location \
    --mypy_out=output/location \
    --grpc_out=output/location \
    --mypy_grpc_out=output/location

Note that generated code for grpc will work only together with code for python and locations should be the same. If you need stubs for grpc internal code we suggest using this package https://pypi.org/project/types-grpcio/

Async GRPC usage

mypy-protobuf generates stubs that are compatible with both sync and async usage, with a few caveats.

In a simple use case, the stubs work as expected.

stub = dummy_Pb2_grpc.DummyServiceStub(grpc.aio.insecure_channel("localhost:1234"))
result = await stub.UnaryUnary(dummy_pb2.DummyRequest(value="cprg"))
typing.assert_type(result, dummy_pb2.DummyReply)

If you need to explicitly type something as an async stub (class attr, etc) then you must use deferred annotations, and the async stub, as it does not exist at runtime.

class TestAttribute:
    stub: "dummy_pb2_grpc.DummyServiceAsyncStub"

    def __init__(self) -> None:
        self.stub = dummy_pb2_grpc.DummyServiceStub(grpc.aio.insecure_channel("localhost:1234"))

    async def test(self) -> None:
        result = await self.stub.UnaryUnary(dummy_pb2.DummyRequest(value="cprg"))
        typing.assert_type(result, dummy_pb2.DummyReply)

_ClearFieldArgType, _WhichOneofArgType_<oneof_name>, _WhichOneofReturnType_<oneof_name> and _HasFieldArgType aliases

Where applicable, type aliases are generated for the arguments to ClearField, WhichOneof and HasField. These can be used to create typed functions for field manipulation:

  from testproto.edition2024_pb2 import Editions2024Test

  def test_hasfield_alias(msg: Editions2024Test, field: "Editions2024Test._HasFieldArgType") -> bool:
      return msg.HasField(field)

  test_hasfield_alias(Editions2024Test(), "legacy")

  def test_whichoneof_alias(
      msg: SimpleProto3,
      oneof: "SimpleProto3._WhichOneofArgType_a_oneof",
  ) -> "SimpleProto3._WhichOneofReturnType_a_oneof | None":
      return msg.WhichOneof(oneof)

  test_whichoneof_alias(SimpleProto3(), "a_oneof")

Note the deferred evaluation (string reference, or from __future__ import annotations. This bypasses the fact that the alias does not exist on the runtime class)

Targeting python2 support

mypy-protobuf's drops support for targeting python2 with version 3.0. If you still need python2 support -

python3 -m pip install mypy_protobuf==2.10
protoc --python_out=output/location --mypy_out=output/location
mypy --target-version=2.7 {files}

Releasing to pypi (TODO: Automate)

  1. Checkout the commit you need to release
  2. Make sure it is tagged with the release version
  3. Make sure the working tree is clean
  4. Clean dist (rm -rf dist)
  5. uv venv
  6. uv pip install --upgrade build
  7. uv run python3 -m build
  8. uv pip install --upgrade twine
  9. uv run twine upload --repository testpypi dist/*
  10. uv run twine upload dist/*

3rd Party Tests

3rd Party proto files can be added in run_tests.sh, they should be cloned in, generated, type checked, then cleaned up.

Contributing

Contributions to the implementation are welcome. Please run tests using ./run_test.sh. Ensure code is formatted using black.

pip3 install black
black .

Contributors

Licence etc.

  1. License: Apache 2.0.
  2. Copyright attribution: Copyright (c) 2022 Nipunn Koorapati