arrow_backAll articles
jiraai developersai coding agent orchestratorclaude code

How to hand Jira tickets to AI coding agents

· Josef Chmel

Keep Jira and add AI developers – what a ticket needs before an AI coding agent can take it, the main ways to hand it over in 2026, and how to keep the merge in human hands.

Short answer: you do not have to leave Jira to give its tickets to AI coding agents. In 2026 you can assign a Jira work item to Atlassian's own coding agent, to Cursor or to GitHub Copilot, or link it to a task in a queue such as mcptask.online, where a runner on your own machine works it. Whichever you pick, the result should be a pull request that passes your tests and that a person decides to merge.

We build mcptask.online, one of the options below, so read that section with this in mind. Every claim about another product links to that product's own documentation or changelog; we checked them in November 2026, and tools in this category change quickly.

What a ticket needs before an AI developer can take it

An AI coding agent works from what the ticket says, not from what the team meant. A ticket it can finish has a small scope: one change that fits into one pull request, not an epic. If you would split the work for a new colleague, split it for an AI developer too.

The ticket names its acceptance criteria. "Login is broken" leaves the agent guessing; "after a password reset the user lands on the dashboard, not on the login form" tells it when it is done and what to test.

The repository has tests the agent can run. Tests are how an AI developer checks its own work before anyone reviews it, and how the reviewer knows the change did not break something else. A ticket in a codebase without tests gets a pull request nobody can trust without reading every line.

The ticket says which repository and, if it matters, which branch. Most of the integrations below either ask for the repository explicitly or pick one from rules you set up in advance.

Your options in 2026

The list covers the main ways to start an AI coding agent from a Jira ticket. It is not a ranking: the right choice depends on where your code lives, where your tests can run and who has to see the work.

Atlassian's coding agent in Jira

Atlassian builds its own agent into Jira. According to Atlassian's documentation, you start the Jira Coding Agent from the Agent sessions section of a work item, or assign the work item to it, pick one or more repositories and select Generate code. It runs in a cloud sandbox, needs Rovo credits, a Standard or Premium plan and Bitbucket Cloud or GitHub Cloud connected to the Jira site, and can push its changes to a draft pull request that your team reviews and merges. It can also be started from an Automation rule. Good for: teams on Jira Cloud that want the agent from the same vendor as the tracker.

Cursor's Jira integration

Cursor's Jira integration starts a Cursor Cloud Agent when you assign a work item to Cursor or mention @Cursor in a comment, optionally with the repository, branch or model. The agent works from the work item's title, description and comments, posts its status and progress to the work item, and links the pull request in its completion update. It requires a Cursor Teams or Enterprise plan with usage-based billing, Jira Commercial Cloud with Rovo enabled, and a repository on GitHub, GitLab, Azure DevOps or Bitbucket connected to Cursor. Good for: teams whose developers already use Cursor.

GitHub Copilot coding agent for Jira

GitHub's Copilot agent can be assigned a Jira issue or mentioned in a comment, with the target repository named in the issue or the comment. It needs Jira Cloud with Rovo activated and a paid Copilot plan, works in the GitHub repository, uses GitHub Actions minutes, and copies the Jira context, including acceptance criteria, into the pull request it opens. Since the general availability release, progress streams to the Jira issue and the agent opens a draft pull request when it finishes. Good for: teams whose code and reviews already live on GitHub.

A ticket queue with a runner on your own machine: mcptask.online

mcptask.online is task management for people and AI developers in one queue. You keep Jira and link a task to a Jira issue; the runner on a machine you dedicate takes the next task and works it with Claude Code, Codex CLI or OpenCode. Good for: teams that want AI developers to run against their own database and tests, on their own hardware, next to the people on the team. The next section describes exactly what it does with Jira.

For a broader comparison of tools that manage several AI coding agents – including Devin, Linear's agents and open-source options – see tools to manage multiple AI coding agents.

Where mcptask.online fits

mcptask.online links tasks to Jira Cloud or self-hosted Jira on all plans; see integrations. The link is per task: start a task name with the issue key, such as PROJ-123, and the issue description is loaded into the task. You can also create a Jira issue from an existing task. The Jira URL and project key are set per project, with a Jira API token per user.

What goes back to Jira is deliberately narrow. When a pull request for the task is opened or merged, a note is added to the Jira issue as a comment, and a merge also adds a "Completed" comment. Work records with their progress can be sent to the issue too. Each of the two is a switch in the project settings. Other messages on the task are not posted to Jira, and statuses are not synced, so your Jira workflow stays as it is.

The work itself happens outside Jira. The runner works on your own machine – a PC or Mac with your project cloned – not in a vendor's cloud sandbox, so it uses your real database and your real tests. It takes a task from the queue, writes the code and tests on a branch, and ends every task as a pull request after the tests and local CI have run. A person on your team decides what happens next: in manual modes the pull request waits for human review, and in auto-squash modes, which you choose, it is merged only after the checks are green. An AI developer is a user at the same price as a person.

If you are weighing whether to move away from Jira altogether, read our Jira alternative for AI agents.

Keeping review in human hands

Whichever option you choose, make the pull request the only way an AI developer's change reaches your main branch. All the options above end in a pull request; none of them needs write access to your main branch to do its job.

Make green checks a condition of the merge. Branch protection that requires your tests and CI to pass stops a change that breaks the build, whoever wrote it. An agent that runs the tests before it opens the pull request saves the reviewer a round trip.

Keep the merge decision with a person. A reviewer reads the change against the ticket's acceptance criteria, and the release stays with your team. When a ticket keeps coming back with the wrong change, rewrite the ticket before you blame the agent.

Frequently asked questions

Do I have to leave Jira to use AI coding agents?

No. Atlassian's coding agent, Cursor and GitHub Copilot all start from a Jira work item, and mcptask.online links its tasks to Jira issues, so your team can keep planning in Jira.

Do AI coding agents merge their own pull requests?

In the options above, the agent opens a pull request and your team decides about the merge. In mcptask.online you choose the mode: a person reviews every pull request, or auto-squash merges it only after the checks are green.

What does mcptask.online write back to a Jira issue?

Comments: a note when a pull request for the task is opened or merged, a "Completed" comment after the merge, and, if you leave it switched on, work records with their progress. Other messages and statuses are not sent to Jira.

Where does the code run?

Atlassian's coding agent and Cursor's cloud agents run in the vendor's cloud, and GitHub Copilot works in GitHub, using GitHub Actions. The mcptask.online runner works on a machine you own, with your project cloned there.

Related articles