Internet Access Whitelist Limits for Kaggle Notebooks

Short answer: Kaggle approves whitelist additions based on security risk, abuse potential, terms-of-service compliance, and demonstrated community demand. There is no public roadmap or timeline for expansion, and there is no mechanism for temporary, per-session access — a domain must be permanently approved before any notebook can reach it.

What Kaggle evaluates

Kaggle has not published a formal scoring rubric, but based on staff responses in Kaggle Meta discussions and support replies, the recurring factors are:

• Security and abuse risk — endpoints that could be used for exfiltration, crypto-mining coordination, or proxying arbitrary traffic are unlikely to be approved.
• Terms-of-service and licensing compliance — the service must permit programmatic access in a way consistent with Kaggle's own rules.
• Community demand — widely used services (major model hubs, public dataset APIs, common cloud endpoints) are prioritized over niche personal servers.

Treat this as the likely explanation rather than a confirmed checklist: the exact internal criteria are not publicly documented, so any claim of a definitive policy should be viewed with caution.

Roadmap and timeline

There is no published roadmap. Additions happen ad hoc after review, and community reports suggest requests can take anywhere from days to several weeks, with no guarantee of approval. Because of this, notebooks intended for reproducible research should be designed to work with currently approved services wherever possible, rather than assuming a pending request will succeed.

Temporary per-session access

No such mechanism exists. The whitelist is global to the platform, not scoped to a user, notebook, or session. If your workflow depends on an unapproved endpoint, the practical workarounds are:

1. Download the required data or model artifacts outside Kaggle and attach them as a Kaggle Dataset, which the notebook can then mount.
2. Restructure the pipeline so the external API call happens in a separate environment (local machine, CI job) and only the results are uploaded.

How to request a new domain

1. Open a discussion on Kaggle Meta or file a support ticket.
2. Include the exact domain, the purpose (e.g., fetching pretrained weights), expected request volume, and why existing approved services cannot substitute.
3. Wait for staff review; do not assume approval when publishing a notebook that depends on it.

Verifying whether a domain is whitelisted

You can test from inside a notebook session:

curl -sS -m 10 -o /dev/null -w "%{http_code}\n" https://api.example.com/health

A timeout, Network is unreachable, or a connection reset generally indicates the domain is not whitelisted; an HTTP status code (even a 4xx) means network egress to that host works. Note that even whitelisted domains remain subject to rate limiting, and traffic patterns flagged as abusive can still be blocked.

Caveat: Kaggle's whitelist policy is not fully documented and may change; verify the current state with a live test and recent Kaggle Meta announcements before relying on any specific endpoint.