Hey @Taiwo-Sh, thanks for the contribution! The overall direction makes sense. I have some feedback after reviewing the code:

Suggestion: use **endpoint_kwargs instead of a premade instance

The main concern with accepting a pre-built WebsocketRPCEndpoint is that PubSubEndpoint needs to register its on_disconnect handler (for subscription cleanup), and the only way to do that on an existing instance is by mutating its private _on_disconnect / _on_connect lists. This creates tight coupling to RPC internals and introduces risk — e.g. the rpc_channel_get_remote_id value can get out of sync between the two objects.

A simpler approach that still solves #96: have PubSubEndpoint forward extra kwargs to WebsocketRPCEndpoint:

def __init__(
    self,
    methods_class=None,
    notifier=None,
    broadcaster=None,
    on_connect=None,
    on_disconnect=None,
    rpc_channel_get_remote_id=False,
    ignore_broadcaster_disconnected=True,
    **endpoint_kwargs,
):
    ...
    self.endpoint = WebsocketRPCEndpoint(
        self.methods,
        on_disconnect=[self.on_disconnect, *(on_disconnect or [])],
        on_connect=on_connect,
        rpc_channel_get_remote_id=rpc_channel_get_remote_id,
        **endpoint_kwargs,
    )

Then the usage from #96 becomes:

endpoint = PubSubEndpoint(
    methods_class=JobFetchRPCMethods,
    broadcaster=settings.REDIS_URL,
    serializing_socket_cls=JobFetchWebSocketProxy,
)

This avoids all the new branching/validation logic, private attribute access, and callback merging — PubSubEndpoint keeps full ownership of the endpoint lifecycle.

Other notes

• The docstring for the endpoint param is truncated (ends with `met)
• setup_server_with_no_methods_endpoint is identical to setup_server_with_premade_endpoint
• test_premade_endpoint_custom_methods_preserved doesn't actually call or verify the custom method
• test_premade_endpoint_with_methods_and_notifier doesn't verify the notifier identity is preserved

Happy to discuss — and thanks again for working on this!