Jupyter Client Connect to Running Kernel via Python

Question

I am trying to programmatically (with Python) interact with my running jupyter kernel as a prototyping experiment.

I have a jupyter notebook running in my browser and I got the connection info from the notebook via the magic command

%connect_info
{
 "signature_scheme": "hmac-sha256",
 "shell_port": 49545,
 "kernel_name": "",
 "iopub_port": 49546,
 "stdin_port": 49547,
 "hb_port": 49549,
 "control_port": 49548,
 "key": "1a359267-f30d84302c39d352d6ac17c3",
 "transport": "tcp",
 "ip": "127.0.0.1"
}

The jupyter_client docs for KernelClient class leads me to believe I can connect with this info and listen to / display code with this object as well as send code to be executed via the kernel I am connected to but, I am not having any luck with what I have been trying such as:

>>> from jupyter_client import KernelClient
>>> connection = {
  "signature_scheme": "hmac-sha256",
  "shell_port": 49545,
  "kernel_name": "",
  "iopub_port": 49546,
  "stdin_port": 49547,
  "hb_port": 49549,
  "control_port": 49548,
  "key": "1a359267-f30d84302c39d352d6ac17c3",
  "transport": "tcp",
  "ip": "127.0.0.1"
}
>>> client = KernelClient(**connection)
>>> client.history()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/path/to/envs/jupyter-nb/lib/python3.5/site-packages/jupyter_client/client.py", line 347, in history
self.shell_channel.send(msg)
  File "/path/to/envs/jupyter-nb/lib/python3.5/site-packages/jupyter_client/client.py", line 143, in shell_channel
socket, self.session, self.ioloop
>>> client.start_channels()
  Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/path/to/jupyter-nb/lib/python3.5/site-packages/jupyter_client/client.py", line 101, in start_channels
self.shell_channel.start()
  File "/path/to/jupyter-nb/lib/python3.5/site-packages/jupyter_client/client.py", line 143, in shell_channel
socket, self.session, self.ioloop
 TypeError: object() takes no parameters
 >>> client.load_connection_info(connection)
 >>> client.execute('print("Hello World")')
   Traceback (most recent call last):
   File "<stdin>", line 1, in <module>
   File "/path/to/jupyter-nb/lib/python3.5/site-packages/jupyter_client/client.py", line 254, in execute
self.shell_channel.send(msg)
   File "/path/to/jupyter-nb/lib/python3.5/site-packages/jupyter_client/client.py", line 143, in shell_channel
socket, self.session, self.ioloop
   TypeError: object() takes no parameters

EDIT: Adding this section for clarity given @Sam H. suggestion I did this but it did not work

>>> from jupyter_client import KernelClient
>>> kc = KernelClient()
>>> kc.load_connection_info(connection)
>>> kc.start_channels()
  Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/path/to/jupyter-nb/lib/python3.5/site-packages/jupyter_client/client.py", line 101, in start_channels
  self.shell_channel.start()
  File "/path/to/jupyter-nb/lib/python3.5/site-packages/jupyter_client/client.py", line 143, in shell_channel
  socket, self.session, self.ioloop
  TypeError: object() takes no parameters

Answer 1

I have played with the KernelClient in the past, with a very little bit of success, thought admittedly very very little success. The only way I have gotten things to work is via a KernelManager (which you did not have). For example try:

from jupyter_client import KernelManager
km = KernelManager()
km.start_kernel()
kc = km.client()
# now execute something in the client
kc.execute("2+2")
while True:
    try:
        kc_msg = kc.get_iopub_msg(timeout=1)
        if 'content' in kc_msg and 'data' in kc_msg['content']:
            print('the kernel produced data {}'.format(kc_msg['content']['data']))
            break        
    except:
        print('timeout kc.get_iopub_msg')
        pass

That typically (but not always) returns:

the kernel produced data {'text/plain': '4'}

I did a quick google search and the only code or SO post I found that had some indication that I was on the right track is: This post.

Note I recognize this is by no means a complete answer but hopefully it's a step in the right direction. If you have had success I would be glad to learn.