Know whether the UI your agent built is actually usable.
AgentScope runs real browsers against complex web apps, detects blocked controls, clipping and responsive failures, explores meaningful states, and returns machine-readable evidence your coding agents can fix.
POST /v2/ui/audit { "target": "https://app.example.com/editor", "options": { "viewport_matrix": "standard", "verify_physical_clicks": true } } critical toolbar overlaps export button major properties panel clipped @ 390px fix_packet src/Toolbar.tsx:184 gate FAIL · exit_code 2
UI QA that reasons about interaction, not pixels alone.
Traditional visual testing asks “did pixels change?” AgentScope asks the harder question: “can a human or agent actually use what shipped?”
Real clickability evidence
Detect a button that exists in the DOM but sits under an overlay, sticky toolbar or higher z-index receiver.
One audit, six viewports
Desktop through small mobile, with findings deduplicated while preserving every affected viewport.
Explore meaningful UI states
Replay tool selection, sidebars, dialogs and editor modes so transient layout failures stop hiding in interaction paths.
Evidence your agent can patch
Selectors, blockers, source hints, component/CSS context and replay steps become a concise machine-readable fix packet.
Block bad UI before deploy
Turn critical overlap, clipping, accessibility and interaction failures into CI exit codes and release policy.
Use from any coding agent
HTTP API and MCP surface mean Claude, ChatGPT, CI workers or your own orchestration can invoke the same runtime.
Build → inspect → fix → prove.
AgentScope fits after implementation and before merge or deployment. It does not replace your coding agent; it gives that agent a better sensory and verification layer.
Audit
Capture geometry, hit stacks, accessibility and screenshots.
Explore
Traverse high-risk editor and application states.
Fix packet
Give the coding agent bounded, source-oriented evidence.
Verify
Re-run only affected states/viewports, then a quick regression.
Pay for useful browser work, not every pixel.
An execution credit tracks an AgentScope browser/UI-QA unit. Start free, then scale concurrency and evidence retention as agents do more work.
Give your coding agents eyes—and a pass/fail standard.
100 execution credits are included. No payment method required for Free.