So you’re running a jupyter notebook on some remote workstation (let’s say that’s firewalled and only allows access via SSH. The notebook server runs on e.g. port 8888.

SSH Port Forwarding

To connect to the server from your laptop, you could use SSH port forwarding:

laptop:~> ssh -L 8888:localhost:8888

You then point the browser on your laptop to http://localhost:8888 to access the notebook.

Reverse Port Forwarding

What, however, if you do this from an iPad, or a locked down Windows computer that doesn’t have SSH or the ability for port forwarding? In this case, you can still get through the firewall around myworkstation by using reverse port forwarding via another non-firewalled (web-)server (the “forwarding server”). Note that exposing a standard notebook server directly is not a good solution for sharing notebooks among a larger group of people on a regular basis. That is something that jupyterhub is designed to solve.

Whenever you expose a jupyter notebook server to the internet, security is a major concern! Anybody who gets access to your notebook server gets a full shell on myworkstation. Therefore, we must

  1. Set a strong password for the notebook server
  2. Use SSL to encrypt the connection. A self-signed certificate will not be sufficient in our case, so we’ll use letsencrypt to get a proper (free) certificate. If SSL is not configured correctly, only connections between the forwarding server and myworkstation will be encrypted (via SSH), while the connection between the laptop/iPad and the forwarding server is completely open, making it easy to break into myworkstation.

To set up the forwarding server, I’m assuming you have a server with root access, and a domain name. I will be using the Digital Ocean server that handles my website ( We need the following configuration steps:

  1. Set up a subdomain in the DNS for the domain, e.g. (pointing to the IP of server). If you’re using something like Cloudflare, the subdomain must not go through their proxy.

  2. Create a new letsencrypt SSL Key for the subdomain: temporarily disable any running webserver (making port 80 available for the letsencrypt authorization procedure), then run e.g.

    server:~> sudo /opt/letsencrypt/certbot-auto certonly –standalone -d

    This does not conflict with the existing certificate for (and having separate keys from my normal website is the entire reason for having set up the jupyter subdomain). The new certificate files end up in /etc/letsencrypt/live/

  3. Configure sshd on server to expose reverse-forwarded ports to the internet. In /etc/ssh/sshd_config, set

    GatewayPorts clientspecified

    Restart the ssh server with service ssh restart.

Now we can run the jupyter notebook server on myworkstation. The certificate files /etc/letsencrypt/live/ and /etc/letsencrypt/live/ need to be copied to some accessible location on myworkstation, e.g. to ~/ and ~/, respectively.

The following command tells the jupyter notebook server to use the key files:

myworkstation:~> jupyter notebook --certfile=$HOME/ --keyfile=$HOME/

Note that you can also set this up permanently in ~/.jupyter/, but since I will only do the forwarding sporadically, I prefer to have it as a command line option on a case-by-case basis.

Lastly, on myworkstation, we make the reverse port forwarding connection to server:

myworkstation:~> ssh server -N -R :8888:localhost:8888

This command must run for as long as we want the notebook server to be accessible.

You could also use the -f option to put this in the background. Note the leading colon in :8888:localhost:8888. This works in conjunction with GateWayPorts clientspecified to ensure that the remote socket listens on all network interfaces, i.e. it will be available via the server public IP, not just via localhost. One could also set GateWayPorts to yes and leave out the leading colon, but the configuration here is much more secure.

The jupyter notebook server will now be accessible from the internet at large at

Reverse Proxying through Nginx

An alternative to the above procedure is to use nginx proxying. This assumes that there is an nginx webserver running on server. The basic idea is this: We start the jupyter notebook on myworkstation, without SSL encryption (but with password protection!). Then, we create a tunnel from server to myworkstation.

server:~> ssh -N -L 8888:localhost:8888

This would make the notebook locally accessible on server (but does not expose it to the internet at large). The last step is to configure nginx to act as a proxy to the locally forwarded port, expose it to the internet, and wrap it in SSL. In the appropriate config file (/etc/nginx/sites-available"/default, in my case), I’ve added a section

server {
    listen 443 ssl;
    ssl_certificate /etc/letsencrypt/live/;
    ssl_certificate_key /etc/letsencrypt/live/;
    ssl_protocols TLSv1 TLSv1.1 TLSv1.2;
    ssl_prefer_server_ciphers on;
    ssl_dhparam /etc/ssl/certs/dhparam.pem;
    ssl_session_timeout 1d;
    ssl_session_cache shared:SSL:50m;
    ssl_stapling on;
    ssl_stapling_verify on;
    add_header Strict-Transport-Security max-age=15768000;
    client_max_body_size 0;
    location / {
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "upgrade";
        proxy_set_header Host $host;

This uses the same certificate files that were created above. Now, upon restarting the webserver (service nginx restart), the notebook server is publicly available at The connection between the browser and server is SSL-encrypted, and the connection from server to myworkstation is encrypted through SSH. The proxy_set_header settings are required to properly forward the websocket connections that the notebook relies on.

The approach has several advantages compared to reverse-forwarding:

  • SSL keys stay on server. Instead of generating a separate key for for, you may just add the jupyter subdomain to the main key that the webserver uses anyway.
  • No need to configure sshd on server to expose reverse-forwarded ports
  • No need to configure SSL for the jupyter notebook server
  • The port number of the jupyter server will not be exposed.

The two drawbacks are:

  • Need to have nginx server running on server
  • Need to interact both with myworkstation (to start the notebook server) and with server (to initiate the port-forwarding SSH connection)


No comments

You may format you comment with Markdown. If your comment is a valid contribution, it may be posted on this page. Your email address will not be made public.