Jupyter Server

Install the package:

pip install jupyter_kernel_client

Install Jupyter server prerequisites:

pip install jupyter-server ipykernel

Start a Jupyter Server with a token:

jupyter server --port 8888 --ServerApp.port_retries 0 --IdentityProvider.token MY_TOKEN

Create a client, execute code, and inspect outputs:

import os
from platform import node

from jupyter_kernel_client import JupyterKernelClient

with JupyterKernelClient(server_url="http://localhost:8888", token="MY_TOKEN") as kernel:
    code = """import os
from platform import node
print(f\"Hey {os.environ.get('USER', 'John Smith')} from {node()}.\")
"""
    reply = kernel.execute(code)
    print(reply)

Typical response:

{
  "execution_count": 1,
  "outputs": [
    {
      "output_type": "stream",
      "name": "stdout",
      "text": "Hey user from host.\n"
    }
  ],
  "status": "ok"
}

You can also manage lifecycle manually:

from jupyter_kernel_client import JupyterKernelClient

kernel = JupyterKernelClient(server_url="http://localhost:8888", token="MY_TOKEN")
kernel.start()
reply = kernel.execute("x = 1 + 1")
print(reply)
kernel.stop()

Connect To An Existing Kernel

If you already have a notebook kernel running in JupyterLab, connect with kernel_id:

from jupyter_kernel_client import JupyterKernelClient

kernel = JupyterKernelClient(
    server_url="http://localhost:8888",
    kernel_id="83ef59b7-9c78-40bd-8cc2-4447635e7d0b",
    token="MY_TOKEN",
)
kernel.start()
reply = kernel.execute("x = 1")
print(reply)

When attaching to an existing shared kernel, avoid shutting it down unless you own the session.

To discover a live kernel_id, start JupyterLab and inspect the active kernel session in the UI or server APIs.