A test recorder, an MCP server & a map that argues back
Hawzu learned to record a test in Chrome and keep it as steps and scripts, opened up to AI clients over MCP, got its pricing, and started telling you when the things describing your product disagree.
- Record a test once in Chrome with the Hawzu Recorder extension, in beta, and the test case keeps it as readable manual steps and two scripts, Playwright (TypeScript) and pytest + Selenium (Python), that your own CI runs, with each result mapping back to the test case by its code.
- The recorder's controls sit in Chrome's side panel, beside the page rather than on top of it, and fields it recognises as passwords, one-time codes or card numbers, or that you mark Secret, are saved only as an environment variable name.
- Write a recording into the test case as its manual steps, replacing or adding after what is there, open the Locators tab to see which locator each script uses and what else was tried, and when a page changes, fix the broken locator once in the project's Object Repository so every test case that uses it gets the new one.
- Connect Claude, Claude Code, ChatGPT, Cursor or VS Code straight to your project over MCP, and ask about your test cases, requirements, defects and releases without leaving the editor you already work in.
- It is read-only by construction and reads exactly what your own role reads, refusing rather than returning an empty list when it may not look. It uses no Hawzu AI credits, works on every plan including the free one, and connects with an access token or by signing in and approving one workspace — administrators can see and disconnect every connection into theirs.
- Atlas now shows you where the things describing your product disagree: your specifications placing a feature somewhere your folder tree does not, parts of the product your tests walk through that the map has never been told about, and two nodes that look like one thing counted twice. Each side sits beside the other, dated and cited, so you can decide which is right.
- A Map tab in Release Analysis and Execution Analysis shows the Atlas map narrowed to what that release, or a single run of it, actually walked, with each journey's health read from that scope, so a sign-off can see which parts of the product were exercised and which were not.
- Duplicate a test case, or a whole folder with its subfolders and every case inside, so a new variant starts from something you already wrote instead of a blank form.
- Comments on defects follow the same rules as comments everywhere else — anyone who can read a defect can read its discussion, you can always edit your own, and removing someone else's takes a moderator.
- Three plans — Free, Growth and Enterprise — with no feature behind any of them: they differ by how much you use, never by what the product can do. Free covers five people in any mix of writers and read-only viewers, with the whole product including the AI, and does not expire, and Growth can be tried free for 14 days without a card, landing back on Free with your work intact.
- Move to Growth yourself in workspace settings — monthly or yearly, priced from your real team before you confirm anything, with read-only viewers at a quarter of a member — alongside a billing card showing the period, the seat cost and every invoice raised so far.