Point ReTibe at your codebase. The agent reads it, writes real scenarios, runs them on a schedule in real browsers, catches visual regressions and security holes, and files the bug before anyone opens a ticket.
Or talk to the agent yourself live demo belowFree plan forever: 20 web tests and 20 bot tests a month. No card.
Playwright MCP drives a live browser, which is exactly right for looking around: open the page, click things, see what is there. Then the session ends and nothing remains. No saved scenario, no report, no history, no schedule. Next week you explain the same flow to the model again. ReTibe reads the page too now — an outline with selectors verified unique — but looking was never the hard part.
They are not competitors, and we say so in our own docs. Playwright MCP is still the tool for driving a browser by hand, and ReTibe no longer needs it to see the page. Keep it if you like it, and hand the flow to ReTibe so it keeps running after you close the laptop.
Our free plan runs 20 web tests a month, forever. The cost of the habit is not money, it is the regression nobody was watching for.
Add the MCP server next to the one you already have and ask the same agent to pin one flow.
See the connectorA progressive L1 to L4 scan pulls routes, forms, auth flows and real selectors into a 300K character context. No guessing at markup it has never seen.
10 to 50 scenarios in minutes, each checked by a validator that catches invented actions and assertions that would pass on anything.
Nightly on a cron, in three engines, with pixel diffs, a security score, an emailed report and a Jira bug when something fails.
Baselines, pixel diffs, an approve and reject queue, and a threshold you control.
TLS, headers, XSS, SQLi, CORS, clickjacking, traversal. Scored, with proof and a fix.
Stability heatmap per step and a 0 to 100 flakiness score, so you know which test is lying.
Cron, daily, weekly or monthly, with colour-coded history and email alerts.
A composite GitHub Action that gates the pipeline on one scenario, JUnit XML for Jenkins and GitLab, and Jira bugs created on failure.
Diff the live page against exported frames and hold the build to the design.
The agent reads the page before it writes a selector for it: headings, forms, buttons, iframes, and selectors verified unique in a real browser.
Two runs diffed step by step, and each step's pass/fail string across the last ten — so a flake and a regression stop looking alike.
The failing step photographs itself and carries the console errors and failed requests from around the moment it died.
Test organisation, run history and scenario recording, in one workspace.
Every run on one timeline, with duration and success trends.
Step-level heatmap, so you see instantly which step fails most often.
Unstable tests found automatically, with a flakiness score from 0 to 100.
Root cause of a failure, with recommendations and a run-to-run comparison.
Enough to see it work on your site
For a team shipping every week
For a QA function with SLAs
Claude connected to ReTibe over MCP, right here on the page. Pick one of the prompts that work, or write your own, and see the tool calls, the validator and the run result.
Too vague. Even an agent that can look at your page cannot guess what matters to you. Give it a URL and say what must be true after each step.
Liked what it did? The same connector runs in your own editor. Issue a key, add one MCP server, and the agent you already use can write and keep your tests.
Get your API key Read the setup guideAutomatic website and app testing in real time
A demo of the console. Free accounts get 20 real runs a month.
No. Keep it for exploration. ReTibe is where a flow goes once you want it checked on every release without you being there.
That is the failure mode we built the validator for. It runs before a browser opens, flags assertions that cannot fail, and there is a /harden prompt that hunts them across your whole suite.
No. Scenarios are JSON, and you can record them in the browser, describe them to the assistant, or let the agent generate them from your codebase.
It looks. retibe_inspect_page opens the URL on the platform and returns the page outline with selector candidates that were verified unique at capture time. A guessed selector is the expensive mistake here: every wrong one burns a run, and an inspection does not — it comes off its own counter, 60 a month even on Free.
Median web test is 67 seconds across real platform history, with the slow tail around eleven minutes. Runs are asynchronous, so nothing blocks while it works.
Twenty runs a month, free forever. Enough to point it at one real flow tonight and read the report in the morning.