Start a project
AI in QA

Claude for QA engineers (Jira and Xray)

Denis de Oliveira
Denis de OliveiraNov 17, 2026
Claude AI for QA Engineers, showing a software tester using Claude with Jira and Xray for test management and Quality Assurance.

Let me start with something every QA engineer knows: most of our work is made of words. User stories, acceptance criteria, bug reports, test cases, logs, release notes. We read them, rewrite them, question them and connect them to each other. A lot of the day goes into that, long before anyone runs a test.

One thing I’ve learned over the years as a QA Engineer is that sometimes the biggest issues are found when we start asking questions that weren’t part of the original requirements.

For example, imagine you’re testing an application where different systems communicate through APIs. Everything seems to be working fine, but what happens if one system updates information and the other doesn’t receive it? What if the API fails halfway through the process? Or what happens if the same request is sent twice?

These are the kinds of scenarios I like to explore, and this is where AI can be really helpful.

I can use AI to review a requirement, challenge my assumptions, and suggest scenarios I might not have considered.

Does AI always get it right? Of course not. Sometimes its suggestions don’t make sense, and that’s where my experience comes in.

For me, AI is like having another person to brainstorm with. It doesn’t replace my knowledge or experience, but it can help me look at a problem from a different angle.

And sometimes, that different perspective is exactly what we need to catch an issue before it becomes a bigger problem.

That is why I wanted to try Claude, an AI assistant made by Anthropic, in my own QA work, and why I am writing this post. I will keep it practical: what Claude can do for a QA engineer, how I would use it every day, and how it can work with two tools many of us use, Jira (where the work is tracked) and Xray (where the tests are managed inside Jira).

One thing I want to be clear about from the start. I do not see AI as something that replaces a tester. Testing is about judgement: knowing what could hurt the user, what the business cannot afford to break, and when a release is not ready. AI cannot take that responsibility. What it can do is take care of the first draft, so that you spend your energy on the parts that need a human.

A quick explanation, for anyone new to this

Claude is an AI assistant you talk to in plain language. You can type a question, paste a document, or attach a file, and it answers or writes something back. It can be wrong, so you always check its work, the same way you would check a new colleague’s work.

Jira is the tool many teams use to track stories, tasks and bugs. Xray is an add-on that lives inside Jira and is used to write test cases, group them into test plans, run them and report the results.

A connector is a bridge that lets Claude work inside another tool, instead of you copying and pasting between them. You may also see the name MCP (Model Context Protocol) for the technology behind it. You do not need to understand the technical side to use it.

The Claude features that matter in daily QA work

It remembers your rules (Projects). Without this, you have to explain your product and your standards again in every chat, which gets tiring fast. In a Claude Project, you save your test strategy, your test case template and your naming rules once. You can also add simple instructions, such as: “Use our template. Never make up features. Mark every guess as ASSUMPTION.” Every new chat in that project starts with those rules already in place. For a QA team, this is how you get consistent results instead of lucky ones.

It reads long documents. Give it a specification, a user story or a log extract, and ask for a short summary, a list of unclear points, or ideas for tests. This is useful when you open a 20-page requirements document and need to understand it quickly. Remove personal data and passwords before you paste anything.

It makes things you can keep (Artifacts). A test plan, a risk table or a checklist can appear next to the chat as its own page. You can edit it and reuse it, instead of scrolling back through a long conversation to find the answer you liked.

It can work inside your tools (Connectors). Atlassian offers an official connector for Jira and Confluence. As Atlassian describes it, you add it in Claude, say yes to the permissions it asks for, and sign in with your Atlassian account. It works with the permissions you already have in Jira, and company admins can decide which AI tools are allowed to connect. One thing to check first: according to a third-party guide, this official connector works with Jira Cloud, not with Jira Data Center. If your company uses Data Center, ask your admin before you plan around it.

It can work inside your test code (Claude Code). Claude Code is a version of Claude for developers and automation engineers. It reads your test project, writes or fixes tests, runs them and shows you what it changed. You can keep your team’s rules in a file called CLAUDE.md, and save repeatable jobs as skills, for example “turn a Jira story into Playwright tests in our style”. If you write automation, this is the feature to look at first.

Four prompts you can try today

A prompt is just the message you send to Claude. The more context you give, the better the answer. Replace the words in square brackets with your own text.

1. Check a story before anyone writes code. This is my favourite use, because a gap found before development costs almost nothing to fix.

Here is a Jira user story and its acceptance criteria: [paste].
List anything unclear or missing, and questions for the product owner.
Do not suggest solutions, only questions.

2. Get test ideas. Use this to widen your thinking, then decide yourself which ideas are worth keeping.

For the story above, suggest test scenarios grouped as happy path,
negative, boundary and accessibility. Use a table with ID, scenario, expected result.
Mark every guess with the word ASSUMPTION.

3. Write a clear bug report. Good for the end of a long day, when your notes are rough but your finding is real.

Turn these rough notes into a bug report with title, steps to reproduce,
expected result, actual result, environment and suggested severity: [paste notes].
Do not add any detail I did not give you.

4. Make sense of a failed test run. Use the answer as a lead to investigate, not as the final verdict.

Here is a failing test log with passwords removed: [paste].
Give me the three most likely causes, from most to least likely,
and what I should check first for each.

The last line in prompts 1 and 3 matters. Telling the assistant not to invent anything makes it much easier to see what it truly knows and what it is guessing.

One area where I find AI really useful is reviewing requirements before testing even begins.

Sometimes a requirement looks clear at first, but once you start thinking about how a real user might interact with the application, questions start coming up.

For example, imagine a feature that allows users to update their personal information. The requirements explain how to save the changes, but what happens if the user leaves a required field empty? What if the internet connection drops while saving? What if two users try to update the same information at the same time?

These are the kinds of questions I can ask AI to help me explore.

Of course, I still need to review its suggestions and decide what makes sense for the application.

For me, the biggest benefit isn’t that AI writes test cases faster. It’s that it helps me think about scenarios I might have missed and ask better questions before development is complete.

And finding those gaps early can save everyone a lot of time later.

How it fits with Jira and Xray

Most Xray teams work in the same way: a story lives in Jira, the tests live in Xray, and the results link back to the story. Here is how Claude can help at each step, with you in charge the whole time.

  1. Read the story. With the Atlassian connector on, ask Claude to fetch the story and run prompt 1. Read the questions it raises before you post any of them to the team.
  2. Draft the tests. Ask for scenarios in your team’s format: step-by-step manual tests, or Gherkin scenarios if you use Cucumber.
  3. Review them like a colleague’s work. Remove duplicates, add the business rules only you know, and check every expected result against the acceptance criteria. This step is the real job. Do not skip it.
  4. Get the approved tests into Xray. There are a few ways, so pick the one your company allows:
    • Xray’s own AI. Xray Cloud has AI Test Case Generation. It turns requirements into manual or Cucumber-style tests, and you review them before the full tests are created.
    • A Marketplace app. For example, Reqase Lite says it supports Claude as an AI provider and lets you approve tests before they go into Xray.
    • Claude Code with an Xray connector. Community-built tools such as xray-mcp say they let Claude Code create and manage tests and executions in Xray Cloud. These are not official Xray or Atlassian products, so ask your security team to review them first.
    • The simple way. Ask Claude for a CSV or a .feature file in the format Xray imports, then import it yourself. Check Xray’s import guide for the exact columns.
  5. Link and run. Link the new tests to the story and add them to a test plan or execution, as you always do.
  6. Summarise the results. After the run, give Claude the results (without passwords) and ask for a short summary and the failures grouped by likely cause. Edit it before you send it to anyone.

If you automate, put your rules in CLAUDE.md or a skill, for example “use our page objects, never use fixed waits, tag every test with its Jira key”. Then ask Claude Code to write the Playwright or Cypress test for an approved scenario, and run it yourself before it joins your suite.

What about the time you save?

I do not want to promise you a number, because I do not know your team, your stories or your tools. What I can say is where the time should come from: first drafts. Drafting scenarios, rewording bug reports, summarising logs and preparing release notes get faster. Deciding what is risky, reviewing the results and signing off do not get faster, and they should not.

Be realistic about the downsides too. In an interview study of 15 testers, saving time was the most common benefit, and wrong or made-up answers were the most common problem. So try it in your own team: measure how long it takes to produce a reviewed set of tests for one story before and after, and count how many AI-drafted tests you had to fix or throw away. Then decide with real numbers.

Simple rules to stay safe

  • Protect data. Use only tools your company has approved, and never paste real customer data, passwords or keys.
  • Mind the permissions. The Atlassian connector acts with your Jira permissions, so use it only where that is acceptable, and read any change before you approve it.
  • Always review. A person reads every generated test and every comment before it is saved or posted.
  • Make sure tests can fail. Change the behaviour a test covers and check that the test fails.
  • Leave a trail. Note in the Jira ticket or test description that a test was drafted with AI and who reviewed it.
  • Be careful with community tools. Treat them like software from a company you do not know yet.

If you want to start this week

Pick one upcoming story. Use Claude only to review the story and suggest scenarios. Review them, put the good ones into Xray yourself, and write down what Claude missed and what it got right. When you trust that routine, add the Jira connector. After that, look at tools for Xray.

After nearly 20 years working in Quality Assurance, I’ve learned that there’s always something new to discover. I’m excited to keep exploring how tools like Claude can make our daily QA work easier, especially when working with Jira and Xray, without losing the critical thinking and experience that make us good testers.

I’d also love to hear from other QA Engineers. Are you using AI in your daily testing activities? What’s working for you, and what challenges have you faced? Let’s share our experiences and learn from each other.

Claude drafts. You decide.