from gen.user_pb import User
Messages are plain Python objects: pass fields as keyword args, or set them later.
user = User(id="123", first_name="Alice")
user.last_name = "Smith"
Serialize to the Protobuf wire format, then parse it back.
data = user.to_binary()
user = User.from_binary(data)
print(user.first_name)  # Alice
print(user.to_json())  # {"id": "123", "firstName": "Alice", "lastName": "Smith"}

Protobuf is the easiest way to build APIs. We recommend using it with Connect for Python, which generates server stubs for you:

from connectrpc.request import RequestContext
from gen.user_connect import UserService, UserServiceASGIApplication
from gen.user_pb import GetUserRequest, GetUserResponse, User
class UserHandler(UserService):
    async def get_user(
        self, request: GetUserRequest, ctx: RequestContext
    ) -> GetUserResponse:
        user = User(id=request.id, first_name="Alice", last_name="Smith")
        return GetUserResponse(user=user)
Serve it with any ASGI server, e.g. uvicorn server:app.
app = UserServiceASGIApplication(UserHandler())

Connect can also generate client libraries for every major language, including your frontend. Here's what the generated Python client looks like:

from gen.user_connect import UserServiceClient
from gen.user_pb import GetUserRequest
client = UserServiceClient("http://localhost:8080")
response = client.get_user(GetUserRequest(id="123"))
print(response.user.first_name)  # Alice

Quickstart

// proto/user.proto
syntax = "proto3";
service UserService {
  rpc GetUser(GetUserRequest) returns (GetUserResponse);
}
message GetUserRequest {
  string id = 1;
}
message GetUserResponse {
  User user = 1;
}
message User {
  string id = 1;
  string first_name = 2;
  string last_name = 3;
}

# buf.gen.yaml
version: v2
inputs:
  - directory: proto
plugins:
  - local: protoc-gen-py
    out: src/gen
  - local: protoc-gen-connectrpc
    out: src/gen

$ uv add protobuf-py connectrpc
$ uv add --dev protoc-gen-py protoc-gen-connectrpc buf-bin
$ uv run -- buf generate

That's all - typed messages and Connect stubs now live in src/gen.

Feature highlights

Generated code you can read

class User(Message[_UserFields]):
    __slots__ = ("id", "first_name", "last_name")
def __init__(
        self, *, id: str = "", first_name: str = "", last_name: str = ""
    ) -> None: ...
id: str
    first_name: str
    last_name: str

Typed oneofs with pattern matching

from protobuf import Oneof
match msg.result:
    case Oneof(field="value", value=v):
        handle_value(v)
    case Oneof(field="error", value=e):
        handle_error(e)

Well-known types with friendly helpers

from datetime import UTC, datetime, timedelta
from protobuf.wkt import Any, Duration, Timestamp
ts = Timestamp.from_datetime(datetime.now(UTC))
td = Duration.from_timedelta(timedelta(minutes=5))
packed = Any.pack(user)

Seamless reflection

# Iterate over message fields
for field in user:
    # Access field name / value
    value = user[field]
    print(field.name, value)
# Check for presence
    print(field in user)
# Assign / delete fields
    user[field] = value
    del user[field]

How it compares

• google-protobuf is complete, but generates code with unreadable binary blobs, broken imports, and a hostile API.
• betterproto improves ergonomics, but it is proto3-only. The original project now points to betterproto2, which calls out active development, incomplete docs, and breaking changes.

Migration guide

| google-protobuf | protobuf-py | | ------------------------------------------ | ------------------------------------- | | msg.SerializeToString() | msg.to_binary() | | MessageType.FromString(data) | MessageType.from_binary(data) | | msg.HasField("nickname") | msg.has_field("nickname") | | msg.WhichOneof("result") + string checks | match msg.result with typed Oneof | | msg.Extensions[ext] | msg[ext] | | msg.child.CopyFrom(other) | msg.child = other |