Hosted Playwright checks
Browser checks your agent can write, run and fix.
Give Claude Code, Cursor or Codex an MCP server. It writes a Playwright check, Flarely runs it, and a failed run comes back as the failing step, the locator, what was expected and what was found. The agent fixes the check. Then Flarely runs it on a schedule and on every deploy.
Free tier: 2 checks, 500 runs a month, no card.
> write a check that checkout completes
on https://shop.example.com
Created "Checkout completes" and ran it.
Run 01M2… failed at step "pay"
getByRole('button', { name: 'Pay now' })
Expected: visible
Received: <element(s) not found>
screenshot · trace
The button says "Place order" on this site.
Updating the check…
Run 01M2… passed in 4.2 s.
Scheduled every 5 minutes, Slack alerts on.Write
Plain Playwright, nothing proprietary
A check is a folder of @playwright/test specs. Push it from the CLI, the dashboard, or let the agent create it over MCP. Every version is kept, and you can clone it out any time.
Run
Trace, video and screenshot on every failure
Runs happen in isolated containers on Cloudflare. Each one records its cost. Failures keep the trace, opened in Playwright's viewer with one click, and a screenshot inline.
Repair
Failure context an agent can act on
Not a stack trace. The test, the step that broke, Playwright's locator, expected and received, the last steps before it, and links to the artifacts. One call, then the fix.
Then keep it running
Schedules and deploy hooks
Cron schedules down to a minute. A hook URL for Vercel, GitHub deployments or Cloudflare Pages that runs every check against the deployed URL. Preview deploys report; production deploys can page.
Incidents and alerts that do not cry wolf
A scheduled failure is re-run once before anyone hears about it. Then Slack, a signed webhook, or email. The next pass resolves it and says so. Agent runs and CI runs never page.
CI and pull requests
A GitHub Action runs the project's checks against the preview URL and leaves one comment: what passed, what failed, the step, and an open-trace link that still works next week.
Teams and tokens
Scoped API tokens for agents and CI, owner, admin and member roles, seats on Team, and an audit trail of who changed what, agents included.
Two minutes to the first check
Create an account, mint a token, add the MCP server to your editor, and ask for a check. Or push a spec folder with the CLI.