This approach sits alongside the SSH method in hoang-quoc-trung/remote-ssh-kaggle-vscode. SSH works well when you can manage keys and open ports, but it can be fiddly on locked-down environments and often pushes you toward extra tooling. Remote Tunnels instead dials out from Kaggle, so there’s no inbound access to arrange and none of the usual ngrok-style headaches (ephemeral URLs, quotas, extra daemons, or port wrangling).
What this notebook does. It starts a lightweight VS Code server inside the running Kaggle session so you can edit, run, and debug from VS Code Desktop. You can find the setup for kaggle in KaggleVSTunnel notebook (It should be the same for Google Colab). The first cell (“bootstrap”) downloads the CLI into a folder under the working directory and extracts it. The second cell (“tunnel”) verifies the CLI, prompts a one-time device login with GitHub or Microsoft, and brings up a named tunnel you connect to from your editor. Keep the tunnel cell running; stopping it closes the connection.
Setup on Kaggle. Enable Internet for the notebook, run the bootstrap cell once per clean environment (when the CLI isn’t downloaded), then run the tunnel cell every session; it will prompt you to sign in—complete the device login in your browser and return to the notebook. If the version check in the tunnel cell fails, simply re-run the bootstrap cell first. Sessions are ephemeral; if your /kaggle/working is persisted and the CLI remains there, you can skip the bootstrap on later sessions.
When you start the tunnel in the notebook, it prints a device-login URL and code—open the link, enter the code, and approve.
Connect from VS Code. In VS Code Desktop, install Remote – Tunnels, open the Remote menu (status bar, lower-left) or the Command Palette, choose Connect to Tunnel…, select the same account provider you used to start the tunnel in notebook, and pick the tunnel name shown by the notebook. Accept the trust prompt and open /kaggle/working so you’re editing the same files the notebook uses.
-
Open the Remote menu and choose Connect to Tunnel…
-
Pick the account provider used to start the tunnel (GitHub or Microsoft).
-
Select the tunnel (e.g., kaggle).
-
Once connected, the integrated terminal runs on Kaggle (And you can check with executing nvidia-smi).
