Oh I see now. You want to run the raw kernel with no front-end on the remote machine and communicate with it via the 0MQ/JSON transport layer. I'm curious, what is the advantage of doing this vs. simply running an instance of Jupyter on a remote machine?