Hi everyone!
We recently hosted the webinar, “From idea to shipped PR: Agents at work in Jira,” showcasing how we’re evolving Jira for AI-native software development. Thanks to everyone who attended!
We didn’t quite get to all the questions, or to the detail we’d like, so we’ve collected the frequently-asked questions and want to provide some answers here, including:
How using agents in Jira compares to using an MCP server
Code repository support
Models and third-party token usage
Details about the Jira agent for Slack
What’s next on the roadmap
Miss the webinar? Watch the recording on demand here: https://www.atlassian.com/webinars/software/agents-at-work-in-jira
Want to learn more about the feature announcement? Read the blog
As of July 2026:
Jira Coding Agent (Jira's built-in agent): Supports Bitbucket Cloud and GitHub.
Cursor for Jira: Supports Bitbucket, GitHub, Azure DevOps, and GitLab.
Claude for Jira (Anthropic): Supports GitHub.
GitHub Copilot for Jira: Supports GitHub.
Codex for Jira (coming soon)
Using Claude as an example, Claude Agent for Jira is built on Claude Managed Agents. This gives enterprises full control over the agent configuration, what tools it has access to, and the environment this agent runs in.
This means that usage is pooled across your organisation, through a single Anthropic API key to run the Claude Agent in Jira.
Tokens and usage reports per session are available in the Claude Console.
See each company’s documentation for exact details:
Microsoft Teams support is coming soon. It's in active development and we expect to share more soon.
To get notified when Teams support ships, follow the Atlassian announcements blog.
Coding agents initiated by Jira run in sandboxed cloud environments, and can be steered from wherever you work (local environments, remove sandboxes, terminals, etc.). Additional details:
Jira Coding Agent runs on Atlassian's Cloud infrastructure. Session data is retained for 30 days and then deleted, is not used for any training, and is under strict access controls.
For third-party coding agents, please see each vendor’s documentation:
Claude for Jira: https://platform.claude.com/docs/en/managed-agents/overview
Cursor for Jira: https://cursor.com/docs/integrations/jira
GitHub Copilot for Jira: https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/integrate-cloud-agent-with-jira
Locally running agents can also write back to Jira using Teamwork Graph CLI or Atlassian’s MCP server.
Yes. Coding agents in Jira check for and read AGENTS.md in the root directory, and are prompted to follow existing patterns in a code repository.
To add additional instructions to agents, you can also link a Confluence "coding standards" page from the work item and the agent will pick it up via the Teamwork Graph.
For Atlassian products:
Jira Coding Agent: Included with paid Jira Cloud editions and requires Rovo usage.
Rovo: Rovo is a separate Atlassian product that harnesses AI to help you find, learn, and act on information throughout your Atlassian and third-party apps. Rovo comes with paid Atlassian Cloud plans, and uses Rovo credits. Learn more:
For third-party agents in Jira:
Claude for Jira: Billed by Anthropic based on the API token you have provided when configuring Claude for Jira. Billing is pooled at the organisation level.
Cursor for Jira: Billed by Cursor against your Cursor Business / Enterprise seat.
GitHub Copilot for Jira: Billed by GitHub under usage-based billing for organizations and enterprises
You do not need to use the agentic template. It's simply a fast-start template that pre-configures columns, statuses, and automation rules for you. You can enable the same capabilities in any existing Jira project by:
Installing the agent marketplace apps you want (Claude for Jira, Cursor for Jira, etc.) at the org level.
Adding a column to your board and assigning an agent to it (Board settings → Columns → Assign agent).
Optionally, creating Jira Automation rules to auto-assign agents based on labels or status transitions.
If the agent template isn't showing up in your list of templates, it may still be rolling out to your instance. Check with your Jira admin, and ensure "AI features" are enabled in the Atlassian admin hub.
Jira Coding Agent is powered by frontier models.
Third-party agents in Jira each use their own model provider.
Other agents, such as agents via AWS Bedrock, Databricks, or local/self-hosted LLMs aren’t currently supported when orchestrated in Jira today. However, these agents can still interact with Jira via Teamwork Graph CLI or the Atlassian MCP server.
Coding agents in Jira get context from Teamwork Graph, Atlassian’s data layer across all your organization’s people, work, and knowledge. This includes a Jira work item’s summary description, other Atlassian apps such as Confluence, relevant information in your linked repositories, and even third-party apps.
Yes, in multiple ways:
Board column trigger: Assign an agent to a specific column (eg “Ready for AI”). Any work item moved into that column is automatically picked up by the agent, with the person who moved it as human co-owner. You can also do this with the workflow editor.
Jira Automation rules: Combine work item-related triggers (such as a label, status change, custom-field value, or even a schedule) with actions to automatically launch coding agents to work on the task. For example, schedule a weekly task for feature flag cleanup or standup prep.
(coming soon) Rovo CLI: Command-line issue creation and agent assignment for scripting.
Multiple repos per work item: Jira Coding Agent and Claude in Jira support multiple repos per work item.
Monorepos: Supported. The agent operates within the repo you link and can navigate multi-package monorepo layouts. Using CLAUDE.md / AGENTS.md to describe your monorepo's structure helps the agent orient faster.
Asking humans for input: When the agent hits ambiguity, it responds inside the agent session. It also respects blockers, if a linked issue is "blocked by" another open item, the agent will wait rather than proceed.
Inline PR comments (Bitbucket/GitHub): Yes — when you leave inline review comments on an agent-generated PR, the agent picks those up when re-invoked to revise the PR. This makes agent PRs feel like a normal human PR review loop.
Attachments & files in Jira work items: The agent reads text-based content in the work item and linked Confluence pages via the Teamwork Graph. Support for reading and uploading various attachment file types is expanding — text formats, such as reading attached logs, are most reliable today.
Non-English languages: The underlying LLMs are multilingual, so you can write work-item descriptions, comments, and Slack invocations in most major languages. However, for best results — especially for technical instructions and code-generation prompts — we recommend English or ensuring your CLAUDE.md/AGENTS.md instructions are clear about your language conventions.
Today, all AI agent features in Jira (eg Jira Coding Agent, Rovo, Claude for Jira,) are Jira Cloud-only today. Jira Data Center customers do have some Atlassian Intelligence capabilities in the DC 10.x release line, but not the full coding-agent lifecycle demonstrated in this webinar.
If you're on Data Center and interested in migrating to Cloud, your Atlassian account team can walk you through migration paths, enterprise controls, and phased rollouts.
Both approaches are valuable. Using coding agents locally is great for individual developer productivity: fast iteration, deep repo context, and tight feedback loops in your own environment.
Orchestrating agents through Jira adds organizational context and automation on top:
Automatic context: The agent inherits the work item, linked Confluence docs, related tickets, and workspace knowledge via Teamwork Graph, no need for copy and paste.
Workflow triggers: Board columns and Jira Automation rules can launch agents automatically (e.g., a PR merges → QA agent generates tests → CI runs).
Visibility at scale: My Agent Sessions shows all your active sessions, filtered by what needs input, what's in progress, what's done.
They also connect. Developers working locally can use the Atlassian MCP server or Teamwork Graph CLI to read from and write back to Jira, so local coding sessions still stay linked to the work item, even when not initiated from Jira.
Jira Coding Agent has completed the external assessment and compliance certifications for SOC 2 and ISO 27001. It doesn’t currently meet HIPAA, BYOK, or data residency compliance.
Other agents in Jira (eg Claude) are not currently supported in HIPAA and FedRAMP instances.
See Atlassian’s AI Trust Center: https://www.atlassian.com/trust/ai
Here are a few more resources that we shared during the Q&A:
Learn about the new Jira: Jira for AI-native software development
More about Claude Agent for Jira: Claude Agent for Jira (Getting Started)
More about Cursor in Jira: Introducing Cursor in Jira
Slack setup: Use Jira Cloud for Slack
Automation: Jira Cloud Automation
Teamwork Graph: atlassian.com/platform/teamwork-graph
Dan Hom
0 comments