Report Builder 8.2 Release Notes: AI Report Building with Rovo and MCP
Report Builder 8.2 for Jira Cloud adds AI report building: you describe the report you need in a chat, and an agent builds it in Report Builder. The agent can run in Atlassian Rovo or in your own AI client connected over MCP (Model Context Protocol).
The AI only assembles the report configuration. Report Builder calculates the figures from live Jira data, the same way as for reports you build by hand.
The AI features are available only in the Cloud (Forge) version of Report Builder, not in Data Center.
Report Builder Assistant: a Rovo agent for Jira reporting
Report Builder 8.2 includes a Rovo agent called Report Builder Assistant. It uses your site’s Rovo, so there is no separate AI subscription or API key to set up.
To open it, click the new AI Assistant button in Report Builder. Rovo Chat opens in a side panel with the agent already selected. On sites where Rovo is turned off, the button is grayed out and the tooltip says why.
Describe what you need, for example, “hours logged per project this month” or “open bugs in MOBL by priority as a pie chart”. The agent finds the projects and people you mention, writes and validates the JQL, and picks the dimensions and measures.
Before the agent creates or changes a report, it shows a short card describing what it’s about to do and waits for your confirmation. When the report is ready, Report Builder shows a notification with an Open button, and the report opens in the same tab next to the chat.
From there you can keep adjusting it in plain language, for example, by splitting it by assignee or limiting it to the current quarter. The report goes into your report list only when you ask the agent to save it.
Requests work in English, German, Ukrainian and Russian, the languages we tested before the release.
Report Builder MCP server for Claude, Cursor and Codex
If your team already works in Claude, Cursor or Codex, you can connect your client to the new Report Builder MCP server and build reports from a chat or your IDE. Any other client that supports remote MCP servers works too.
Each user connects their own client on the new Configure AI MCP page in the Report Builder menu. Jira access is granted through Atlassian’s OAuth consent screen, and the page then generates a personal MCP token and a configuration snippet to paste into the client. This is done once per user.
Reports built over MCP start as drafts with a preview link. When you ask for a change, the agent edits the same draft, so the link stays valid. Drafts that nobody saves are deleted after 24 hours.
Through the MCP server the agent can:
- search and open reports already saved on your site, so an existing report can be reused instead of rebuilt
- use templates from the Report Builder gallery
- look up Jira projects, epics, issues, users, boards, sprints, components and versions by name
- validate JQL and get the matching issue count
- read the dimensions, measures and filters available on your site, including custom fields
- create, update and save report drafts
AI reports you can build in Jira
Rovo and MCP use the same tools on the Report Builder side, so they can build the same reports.
Universal Visual Builder reports
Single numbers, breakdowns, pivot tables, issue lists and charts over time. A few requests that work well: “issue count by assignee and issue type across MOBL and DESIGN”, “the 20 oldest unresolved bugs with created date and assignee”, “time logged on MOBL last quarter by person”.
Gallery templates
When a question matches a built-in template, the agent uses it instead of building a report from scratch. The templates include Workload, Timesheet, Time in Status, Created vs Resolved, Velocity, Burnup/Burndown, First Reply Time, Lead Time, Ticket Bouncing and Calendar.
Scripted Reports
The agent can also write a Scripted Report, an HTML and JavaScript page rendered over your Jira data. This is how you get reports a pivot table can’t produce: a support dashboard with charts and KPI tiles on one page, a heatmap of issue creation by weekday and hour, a Mermaid dependency graph for an epic, or an invoice built from worklogs.
The Scripted Reports Creator permission still applies. Users who don’t have it can’t create scripted reports through the AI either.
Data access and security
- The agent works under the user’s own Jira permissions and sees only what that user can see.
- Jira access is read-only. The agent can’t change issues in any way.
- Report Builder doesn’t store copies of Jira issues, worklogs or comments.
- For MCP, the OAuth refresh token is stored encrypted and the MCP token only as a hash. Clicking Disconnect on the Configure AI MCP page revokes access.
- If an agent sends a configuration with a field that doesn’t exist on your site, Report Builder rejects it with a validation error, so broken reports don’t end up in your list.
Availability of Report Builder 8.2
Report Builder 8.2 is available for Jira Cloud on the Atlassian Marketplace. The Report Builder Assistant needs Atlassian Rovo to be enabled on your site.
If you have questions or feedback about the AI features, let us know on our support portal.