AmazonCreatorsApi has no proxy support — regular calls and OAuth2 token refresh both bypass HTTP proxy

[mekarpeles]

Note

I liberally used AI Models to help with the organization of this issue; it's a real use case motivated by Open Library (which requires http_proxy in order to make network requests)

Summary

Two independent gaps prevent AmazonCreatorsApi from routing traffic through an HTTP proxy.

Problem 1 — no proxy parameter on AmazonCreatorsApi

AmazonCreatorsApi.__init__ creates ApiClient without a Configuration, so
configuration.proxy is always None. RESTClientObject already handles urllib3.ProxyManager
when configuration.proxy is set — the plumbing is there, it just isn't reachable from the
public API.

Problem 2 — OAuth2TokenManager.refresh_token() bypasses proxy entirely

Even if configuration.proxy were set, token refresh would still go direct:

# oauth2_token_manager.py
response = requests.post(self.config.get_cognito_endpoint(), ...)

This bare requests.post() has no proxy configuration. Token refresh on a cold cache
(worker restart, token expiry) cannot be routed through an authenticated proxy.

Real-world context

Open Library (Internet Archive) runs behind an authenticated Squid proxy. Adding Creators API
support required two separate workarounds to paper over both gaps:

• feat: migrate Amazon PA-API to Creators API (side-by-side) internetarchive/openlibrary#12364 — injects proxy config into rest_client after construction
• feat(proxy): add per-service proxy auth via get_proxy_params() internetarchive/openlibrary#12724 — dynamically subclasses OAuth2TokenManager to override
  refresh_token() with a proxy-aware requests.Session

Both hacks could be removed if proxy support were native to the library.

Proposed fix

Add proxy: str | None = None to AmazonCreatorsApi.__init__ (e.g. "http://user:pass@proxy:3128"),
pass it to ApiClient via a Configuration object, and have OAuth2TokenManager use a
requests.Session with .proxies set when a proxy URL is provided.

The Configuration class and RESTClientObject already have the right structure — this is
purely a wiring change.

Will proactively submit a PR for your consideration