Do not spin up a stateful session for a one-shot scrape - the /scrape, /screenshot, and /pdf quick-action endpoints exist for exactly that and skip session management entirely. Save sessions for multi-step, stateful work where you need cookies and control. And if you self-host steel-browser, set baseURL (Node) or base_url (Python) on the SDK - the default points at Steel Cloud, so a forgotten override sends your traffic to the wrong place.

Context: Steel browser README: two ways to interact - Sessions for stateful workflows with fine-grained control, and the Quick Actions API (/scrape, /screenshot, /pdf) for simple read-only on-demand jobs. The Node and Python SDKs work against both Steel Cloud and self-hosted steel-browser instances, switching via the baseURL / base_url option.