It's a compatibility layer for projects already built on grpcio. The generated clients and servicers plug into your existing gRPC channels and servers, but serialize protobuf-py messages directly. This lets you upgrade to a better Protobuf package without touching your RPC stack.

Quickstart

Generated code requires the runtime libraries protobuf-py and grpcio. The plugin works with Protobuf compilers like buf.

$ uv add protobuf-py grpcio
$ uv add --dev protoc-gen-py protoc-gen-grpc-py buf-bin

Add protoc-gen-grpc-py alongside protoc-gen-py in your buf.gen.yaml:

version: v2
inputs:
  - directory: proto
plugins:
  # Generates message types (*_pb.py).
  - local: protoc-gen-py
    out: src/gen
  # Generates gRPC service stubs (*_pb_grpc.py).
  - local: protoc-gen-grpc-py
    out: src/gen

To generate code for all Protobuf files within your project, run:

$ uv run -- buf generate

A *_pb_grpc.py file is generated for each proto file that declares a service. Message types come from protoc-gen-py, which also manages __init__.py files.

Feature highlights

Well-typed clients you can read

For each service, the plugin emits a client whose methods are fully typed, so editors and type checkers understand every RPC method, its request and response types, and its streaming shape:

async with grpc.aio.insecure_channel("localhost:50051") as channel:
    client = ElizaServiceClient(channel)
    response = await client.say(SayRequest(sentence="Hello!"))
    print(response.sentence)

Servicers that register themselves

Each service also gets a servicer base class with typed method signatures and an add_to_server method:

class ElizaService(ElizaServiceServicer):
    async def say(
        self, request: SayRequest, context: grpc.aio.ServicerContext
    ) -> SayResponse:
        return SayResponse(sentence=f"You said: {request.sentence}")


async def serve() -> None:
    server = grpc.aio.server()
    ElizaService().add_to_server(server)
    server.add_insecure_port("[::]:50051")
    await server.start()
    await server.wait_for_termination()

asyncio and sync, side by side

Every client and servicer is generated in both asyncio (grpc.aio) and synchronous flavors:

with grpc.insecure_channel("localhost:50051") as channel:
    client = ElizaServiceClientSync(channel)  # synchronous servicer
    response = client.say(SayRequest(sentence="Hello!"))

protobuf-py messages end to end

Requests and responses are protobuf-py messages. They have readable generated code, typed oneofs with pattern matching, real IntEnum enums, and a high-performance Rust encoder/decoder, with zero dependency on the legacy google-protobuf runtime.

$ uv run protoc --proto_path proto \
    --py_out src/gen \
    --grpc-py_out src/gen \
    proto/a.proto proto/b.proto proto/c.proto

Example

See the example for a complete client and server using the generated stubs, including server, client, and bidirectional streaming.