I’m looking for a clear methodology or tool (or an MCP) that can help with end-to-end testing of a system, covering both front-end and back-end.
My main focus at the moment is front-end testing, as the system contains a large number of pages, internal features, and complex user flows. This makes manual testing very time-consuming, exhausting, and prone to missing edge cases or unexpected behaviors.
Are there any practical tools or solutions that help with:
Automated UI / front-end testing
Verifying user flows across multiple pages
Detecting bugs or unexpected behaviors
Reducing reliance on manual testing as much as possible
Tools or MCPs that integrate well with Cursor
Any real-world experience, tool recommendations, or best practices for testing large and complex systems would be greatly appreciated
For frontend E2E testing, the standard options are:
Automation tools:
Playwright: a great choice for UI tests, supports many browsers, and has a powerful API for interacting with elements
Cypress: a popular alternative with a nice dev experience, great for testing complex flows
Selenium: older, but works everywhere
Cursor Agent + Browser Tools:
Cursor has built-in browser automation tools that the agent can use to interact with web apps. This can help generate test scenarios or debug.
MCP for deeper integration:
If you need integration with external systems (CI/CD, reporting, test data databases), you can set up MCP servers: Model Context Protocol (MCP) | Cursor Docs
Are you on Mac or Windows, and is this a website or something else?
There are a large number of “basic” web tools, but if you want to skip directly to “full power”, try aurafriday.com - the chrome extension part of that is 10x more powerful than anything else out there, and the system and terminal tools give the AI 100% control over every aspect of your test machine (run this in a VM) - the desktop, all menu options and otherwise “blocked from inside” features etc: gives you the power to test exactly “as if a real person” no matter what you’re testing.
Unless you’re a TDD junkie you’re best off writing some markdown files that describe the aspects you want to test, and use the Playwright MCP to test a few happy paths in plain English.
Lots of examples out there in the wild for this.
I tend to create some rest tenancies and users and store details in a playwright.md file I can dump into the agent chat and add to my gitignore
Be careful with it, sometimes it hallucinates and tries to login to random sites that are not staging !!! I believe there’s MCP protection available in cursor settings which mitigate this these days but early on don’t leave it to its own devices
The system I want to test is a relatively large HR web system that contains many pages, internal features, and a large number of buttons and user flows.
It is a web-based system, and I’m working on Windows.
My main challenge is reducing manual testing as much as possible, especially for front-end flows that span multiple pages and roles (HR, employees, admins, etc.).
I’d appreciate any concrete recommendations or real-world setups that work well for large HR systems.
The system I want to test is a relatively large HR web system that contains many pages, internal features, and a large number of buttons and user flows.
It is a web-based system, and I’m working on Windows.
My main challenge is reducing manual testing as much as possible, especially for front-end flows that span multiple pages and roles (HR, employees, admins, etc.).
I’d appreciate any concrete recommendations or real-world setups that work well for large HR systems. @Chris_Drake@Peter_Cox@deanrie
Although there’s many great options mentioned in this thread, in your case (HR tools) i’d let an agent inside Cursor do unit tests as you go. Then test extensively yourself, as end user. Production testing would make your software more solid, as HR tools can’t permit any bugs. One small detail can make a huge impact on the company using your tool.
What @Visionary said is #1 - everything should be exhaustively tested during the build; it would be more like verification at the end, not testing.
I do crazy-complex web stuff on all kinds of sites very regularly, using mcp-link and it never lets me down.
That server comes with a python tool which can make mcp calls directly - so it’s very efficient to tell the AI to code something in python that uses the browser mcp tool to do the tests (if that’s a possible scenario for you - probably is - testing all the limits and edge-cases it lots of repetition) - writing and running one test program one time is WAY cheaper and faster (and can me more exhaustive) than making the AI manually do repetitive tests itself.
Plus, if you keep the python it writes - you can re-run the tests without any cost or AI later too.