🚀 Launch your local AI coding agents, straight from Jira

[#1]

How many times a day do you copy-paste from Jira into your AI coding tool?

What if that number was zero?

You can now launch your AI coding agent directly from any Jira work item, context included.

 

❓ How does it work? One click from the Development panel in any Jira work item. Pick your coding tool, and it opens pre-loaded with your Jira context. Zero copy-paste.

🛠️ What tools are supported? At launch, Claude Code, Cursor, GitHub Copilot, OpenAI Codex, VS Code, and Rovo CLI - with more coming soon.

🧠 How does it know what I’m working on? Atlassian MCP feeds your agent the work item summary + description automatically, and it can pull more from Jira and Confluence as it goes.

How to Get Started

Requirements

Before you get started, make sure you have:

• A Jira Cloud instance (Free, Standard, Premium, or Enterprise)

• Your preferred AI coding tool installed locally (e.g., Claude Code, Cursor, VS Code with GitHub Copilot, in terminal)

• Atlassian MCP configured for richer context (optional but recommended)

Steps

1. Open any Jira work item and navigate to the Development panel (the same place you manage branches and PRs).

2. Click "Open in coding tool" and select your agent from the dropdown

3. For desktop apps (Cursor, GitHub Copilot, Codex): Your agent opens with the work item context pre-filled. Press enter and start building.

4. For terminal/CLI tools (Claude Code, Rovo Dev CLI): A modal appears with the full prompt ready to copy into your terminal in one action.

For more information, check out the support article.

---

Have questions? Drop a comment below – we'd love to hear from you.

If you’re interested in shaping the future of agents in Jira or sharing any feedback to make Jira work better for you and your team, we’d love to hear.

Book a time to chat with the team 🗓️. We’re excited to hear what you think!

[Jason Krewson]

This is really cool, thanks for sharing!

I will test this out with Cursor and see how it goes. 

[Jason Krewson]

@Himanshu Singh ,

I was testing this today because I'd like to use the new Open in coding tool feature with Cursor.

When I open a Jira work item and click Connect development tools I'm taken to the Development setup page because I don't currently have any development tools configured. After clicking Finish setup the available tools I see are things like GitLab, Bitbucket, Azure DevOps, Jenkins, Octopus Deploy, Snyk, etc.

What I do not see are any of the AI coding tools mentioned in the announcement, such as Cursor, Claude Code, GitHub Copilot, Codex, VS Code, or Rovo CLI.

The documentation says that connecting the Atlassian Rovo MCP server is optional, while installing a supported coding tool locally is required (Cursor is installed on my laptop). My understanding is that Cursor should therefore be available without first configuring MCP.

Am I looking in the wrong place or is this feature being rolled out gradually?

I also reviewed this, Open a Jira work item in an AI coding tool , but it didn't help in this situation. 

I also used the "Give us feedback" in the Development tab and mentioned the same.

[Chris Collins]

We were excited to receive the notification about the new Jira AI coding tool deeplink feature and have had a chance to assess it. It is addressing a real problem space that we have also encountered and spent effort to solve; how to automate an AI-assisted development session initiation from Jira. 

The outcome of our assessment is that we do not expect to adopt the deeplinks feature in it's current form. Let me describe our current tooling flow to explain why.

In our workflow, moving a Story to In Progress triggers our tooling. Our tooling then;

- Resolves the target repository from Jira metadata.

- Claims the issue to be worked on a single specific workstation.

- Sets up the local sandbox in which to work the ticket. This includes;

  - Creating a Jira-linked branch and worktree.

  - Defining non-secret session environment variables and loading required secrets from our secret manager into child processes.

- Only then do we open a new Codex session in that prepared worktree. We also seed that session with our standard initial prompt.

The deeplinks feature appears to solve the context handoff problem, which is useful, but it does not help with the local execution setup problem. Moreover, by opening the AI tool directly, outside our prepared sandbox, it introduces a security concern: the AI tool may be started with a broader or less controlled local context than our best practices permit.

From our perspective, the safe unit of work is not simply “this Jira issue”, it is “this Jira issue, in this repository, on this branch, in this worktree, with this scoped environment.”

We may be more conservative than other teams, and we would be interested to hear how other teams approach this problem and whether they intend to adopt the deeplinks feature. We just wanted to share our use case because the feature is close to a workflow we care about.

[Himanshu Singh]

@Jason Krewson Thanks for checking out deep linking and glad you are excited about it! Since the feature is gradually rolling out it may not have landed for your site just yet.

Feel free to email me at hsingh10@atlassian.com with your site details, and I'll work with the team to get you access earlier.

[Jason Krewson]

Thanks @Himanshu Singh , I sent you an email.

[Himanshu Singh]

@Chris Collins Thanks for sharing your team's workflow with agents, it’s really helpful feedback as we start to plan the evolution of this feature.

You’re right that the current deeplink experience is focused on just the Jira work item context handoff to agent. It seems your workflow also requires deciding the right repo, preparing the branch, setting up the local environment, and then launching your agent with the right prompt.

These are areas we’re exploring, including more flexibility to customize the initial prompt and greater control over how the handoff into a coding tool happens.

We'd love to understand your workflow better to see how we can align it better to what matters to your team.

Would you be open to booking time with the team and how we can make deeplinking work better for your use-cases?