← screencli

What is screencli

screencli verifies AI-coded changes in a real browser — an AI review on every pull request, plus a CLI you can run from any terminal or coding agent.

screencli opens your app in a real browser and checks that a change actually works — not that the code compiles, but that a person can use it. There are two ways to run it:

Both produce the same thing: a browser recording and a plain-language verdict, viewable on a shareable link at screencli.sh/v/<id>.

Why a browser check

Your CI can be fully green while the feature is still broken. Type-checks confirm the types line up, unit tests confirm the reducer returns the right value, and a preview deploy confirms the URL returns a 200 — none of them click the new button. screencli is the layer that opens the running app, performs the change's user-facing flow, and reports whether it worked.

Pick your path

How the GitHub App verifies a PR

When a pull request's preview deployment goes live, screencli:

  1. Resolves the repo config and the preview URL, then dispatches one isolated cloud sandbox for the PR.
  2. Inside the sandbox, clones the repo at the PR's head commit and fetches the unified diff.
  3. Runs a headless coding agent that reads the diff, writes a single Playwright test covering the change's user-facing behavior, and runs it against the preview URL — recording the whole session as video.
  4. Posts (or updates) a PR comment with the overall verdict and a recording thumbnail that links to the watch page.

The recording and verdict also appear on your dashboard.

All pages

Get started with the GitHub App
Connect the screencli GitHub App, pick the repos it watches, optionally configure the preview URL and login, then open a PR to get an AI browser test and a verdict comment.
Configure a repo
Per-repo settings for the GitHub App — turn testing on or off, add baseline smoke tests, override the preview URL, bypass a protected preview, and sign the agent in before it tests a logged-in flow.
Use the CLI
Record, verify, and export browser sessions from any terminal or AI coding agent with npx screencli — no GitHub App required.
CLI login & API access
Sign the screencli CLI in with `npx screencli login`, reuse the token in CI with `npx screencli token`, and call the screencli API directly with your `sc` key.
Verify with your coding agent
Wire `npx screencli verify` into an AI coding agent's inner loop — Claude Code, Cursor, Cline, Windsurf — so it browser-checks its own change and gets a pass/fail verdict before it opens a PR.
The dashboard
Record a browser session straight from screencli.sh — describe what to record, screencli drives the browser in the cloud and gives you a share-ready video and a pass/fail verdict, all listed on your dashboard.
Writing effective prompts
The prompt is the assertion screencli checks — describe the observable outcome you want confirmed. How to phrase prompts for the dashboard composer, the CLI, and baseline tests so the agent verifies the right thing.
Recordings & sharing
Understand a recording's verdict and metadata, control who can see it, and share it — a public watch link at screencli.sh/v/<id>, an embeddable iframe, or a downloadable MP4.
Plans, credits & billing
How screencli billing works — the Pro plan, the shared credit pool that meters agent runs, bringing your own Anthropic key to run unmetered, and managing your subscription.
Team & workspaces
Invite teammates to your screencli workspace so they share one Pro plan, one credit pool, and all of the workspace's recordings.
Troubleshooting
Fixes for the common cases — screencli didn't run on your PR, the verdict came back inconclusive, the test couldn't sign in, the wrong URL got tested, or you ran out of credits.