Runtime architecture · secure browser automation

OpenClaw, Playwright/WebWright, and Cloudflare

Cloudflare protects the public application edge and can also protect the target site. OpenClaw orchestrates the propgent agent; a controlled browser worker performs the site interaction and streams progress back through the backend.

OpenClaw scraper Cloudflare architectureA left-to-right request flow from the Propgent application through Cloudflare to OpenClaw, then to the Propgent agent and a Playwright or WebWright browser worker, which reaches a target website behind its own Cloudflare boundary. Results return as server-sent events. APPLICATION PUBLIC EDGE OPENCLAW RUNTIME TARGET SITE HTTPS / SSE POST /v1/responses browser task HTML / data SSE deltas to client Propgent app backendbrowser UI · auth · SSE fan-outstable user/session id CloudflareDNS · WAF · Accessrate limits · TLSpublic edge OpenClawGateway :18789/v1/responsesstream: true propgent agenttool policy · prompttask/session context Browser workerPlaywright / WebWrightisolated context · cookiesscreenshots · DOM · APIs Target websiteapplication · data · loginbehind its own Cloudflare enforce at edge Cloudflare is a boundary, not the scraperProtect your gateway; treat the target site's controls as authoritative. FLOWrequest / APIresult / stream BOUNDARYCloudflare / trust zone

1 · Inbound request

The app sends an authenticated request to the public Cloudflare hostname. Cloudflare terminates TLS and applies WAF, Access, bot/rate policies, and origin restrictions before forwarding to OpenClaw.

2 · Agent orchestration

OpenClaw receives POST /v1/responses, routes it to propgent, and the agent decides when to invoke the browser tool. Keep the gateway token server-side.

3 · Browser + return

Playwright/WebWright runs in an isolated worker, visits the target normally, and returns structured findings. OpenClaw emits SSE deltas back through the app to the browser UI.

Operational guardrails

  • Use stable session/user IDs for continuity.
  • Apply timeouts, concurrency caps, retries, and per-domain rate limits.
  • Respect the target site's terms, robots guidance, authentication rules, and Cloudflare challenge decisions.

What Cloudflare does not do

  • It does not make a scraper authorized to access a site.
  • It is not a replacement for browser isolation or agent policy.
  • A target site's Cloudflare boundary remains separate from your gateway's boundary.
OPENCLAW · PROPGENT · PLAYWRIGHT / WEBWRIGHT · CLOUDFLARE · ARCHITECTURE OVERVIEW