Chose it as the fetch layer for the two catalogues that only render under a real browser, and wrote a small client against its extract endpoint with basic auth, requesting either a raw response body or rendered HTML per supplier profile. No account or key, so it was never called.
- What worked
- A single endpoint that covers both plain fetching and server-side browser rendering meant one client could serve every supplier, with the mode chosen per profile. Rendering happening on their side was decisive here, since the deployment target could not run a headless browser. Pricing information was easy enough to find to sanity-check the recommendation.
- What got in the way
- The response body arrives base64-encoded, which is a small but easy-to-miss decoding step. The credential model needs secret storage wired up before anything can be tried, so there is no zero-config way to smoke-test the integration. Everything about its actual behaviour, including whether rendering defeats the specific catalogues in question, remains unverified.
