# AI and Jira.

URL: https://levystreet.com/integrations/jira/
Updated: 2026-10-19

An AI Jira integration guide for NZ: Rovo credits and overage billing from 3 December 2026, the Rovo MCP server, REST API, webhooks and triage designs.

## Short answer

An AI Jira integration uses a model to triage incoming issues, flag likely duplicates and prepare engineering handoffs, through Atlassian Rovo, the Rovo MCP server, or the Jira REST API and webhooks. For New Zealand teams the 2026 change to watch is cost: Rovo credits are pooled per user, MCP calls spend them, and extra usage billing switches on by default on 3 December 2026.

- Jira Standard includes 25 Rovo credits per user monthly, Premium 70 and Enterprise 150.
- From 3 December 2026 extra Rovo usage bills at $0.01 per credit unless an admin caps it.
- The Rovo MCP server is free to connect, but every call consumes Rovo credits.
- REST API v3 expects Atlassian Document Format for descriptions and comments.
- Dynamic webhooks expire 30 days after creation unless the app refreshes them.

## What Rovo does natively (October 2026)

Rovo (Search, Chat, Agents and Studio) comes with every paid Jira cloud plan, metered in Rovo credits. Per [Atlassian's usage limits page](https://support.atlassian.com/rovo/docs/rovo-usage-limits/) on 19 October 2026, Jira and Confluence include 25 credits per user per month on Standard, 70 on Premium and 150 on Enterprise; Service and Teamwork Collections include 250, 700 or 1,500. Credits pool across the organisation and do not roll over. Basic AI interactions use 10 credits, and Atlassian says typical prompts use 10 to 25.

On 3 December 2026 extra usage billing starts, switched on by default, at $0.01 per credit (the page does not state a currency). A worked example: a 20-person team on Standard has a 500-credit pool. A triage agent touching 300 new issues a month at about 15 credits each uses 4,500 credits, so roughly 4,000 extra credits, or $40 a month at list price, unless an admin caps it.

## Integration routes and when to use each

| Route | What it gives you | Use it when |
|---|---|---|
| Rovo Search, Chat and Agents | AI across Jira and Confluence, billed in Rovo credits | Usage is modest and the team is on Standard or above |
| Rovo MCP server v2 | `mcp.atlassian.com/v2/mcp` with OAuth 2.1 and each user's permissions; spends Rovo credits | Engineers want Jira inside Claude, ChatGPT or a coding agent |
| Jira Cloud REST API v3 | Issues, comments, transitions and JQL search | You run triage with your own model, billed on your model account |
| Webhooks | `jira:issue_created`, `jira:issue_updated`, `comment_created` and more, filtered by JQL | Triage should start when an issue arrives |
| Forge apps | Atlassian's app platform; new extensibility features are Forge-only | You need UI inside Jira |

## Automation 1: triage for incoming issues


- **Inputs**: New issues in the support or bug project, delivered by a JQL-filtered webhook: summary, description, environment and attachment names.
- **Model step**: Suggest a component and severity against your written definitions, list up to three possible duplicates from a JQL text search with reasons, and draft questions for missing information.
- **Human approval**: The triage lead sees suggestions in an internal comment and applies or rejects them. Priority, assignee and status are never changed by the model.
- **Write-back**: An internal comment, an `ai-triaged` label and an "AI severity suggestion" field, so accuracy can be measured against the human decision.
- **Failure and recovery**: The `X-Atlassian-Webhook-Identifier` header removes duplicate deliveries. A 15-minute JQL sweep for untriaged new issues catches anything missed.


## Automation 2: an engineering handoff that engineers accept


1. **Trigger.** A service issue moves to "Escalated to engineering".
2. **Gather.** Read the description, comments, linked issues and impact fields.
3. **Draft.** The model writes observed and expected behaviour, steps to reproduce, affected customers and open questions, using only what the record says and marking unknowns.
4. **Approve.** The support lead edits the draft, then creates the linked engineering issue.
5. **Learn.** If the engineer bounces it, their reason is stored and becomes evaluation data.


Our [AI agent development](/services/ai-agent-development/) work treats the bounce rate as the operating number for this workflow.

## Permissions and approval design

The [Rovo MCP server](https://support.atlassian.com/atlassian-ai-gateway/docs/get-started-with-the-atlassian-remote-mcp-server/) acts with the user's existing permissions across Jira, Confluence and Bitbucket. Atlassian advises least privilege, reviewing high-impact changes before confirming, and watching audit logs. For custom triage, use an OAuth 2.0 (3LO) app with only the scopes it needs, or a dedicated bot account; Atlassian reserves basic auth with API tokens for personal scripts and bots.

Customer-submitted issues are untrusted input. Prompt injection is LLM01 in the [OWASP Top 10 for LLM Applications 2026](https://genai.owasp.org/resource/owasp-genai-llm-top-10-2026/), so the triage model gets no tools that change issues; see [prompt injection testing](/engineering/prompt-injection-testing/). Issues also carry customer names and log extracts, so strip personal information the model does not need.

## Costs and limits

- **Rovo**: as above; billing admins see usage by app and user under Insights, Platform usage.
- **API quotas**: Atlassian describes three simultaneous limits for apps: an hourly points quota (a default 65,000-point global pool per app), per-second burst limits and per-issue write limits. API-token traffic sits outside the points model but under burst limits.
- **Own-model triage, worked example**: 300 issues a month at 6,000 input and 600 output tokens on Claude Sonnet 5.5 is about 300 × US$0.018 = US$5.40, billed by your model provider rather than from the Rovo pool.

## Where it goes wrong

- **Silent overage.** Extra usage is on by default from 3 December 2026. Set a cap now.
- **Formatting.** Version 3 uses [Atlassian Document Format](https://developer.atlassian.com/cloud/jira/platform/rest/v3/intro/) for descriptions, comments and multi-line fields; single-line text fields take plain strings.
- **Expiring webhooks.** Dynamic webhooks expire after 30 days unless refreshed, OAuth apps get five per user per tenant, and Atlassian says delivery is not guaranteed.
- **Over-eager duplicates.** Similar is not identical. Suggest links; never auto-close.

Support teams often triage in a help desk first; see our [Zendesk](/integrations/zendesk/) and [Slack](/integrations/slack/) guides, and [MCP for business systems](/engineering/mcp-for-business-systems/) for gateway design.

## Questions people ask

### Does Atlassian charge for Rovo in Jira?

Rovo is included in paid Jira cloud plans through monthly credits: 25 per user on Standard, 70 on Premium and 150 on Enterprise, pooled across the organisation and not rolled over. Rovo is not available on Free and needs a verified business domain. From 3 December 2026, usage beyond the pool is billed at $0.01 per credit unless an admin sets a cap or turns extra usage off.

### Can Claude or ChatGPT work with our Jira issues?

Yes, through the Atlassian Rovo MCP server at https://mcp.atlassian.com/v2/mcp, which supports ChatGPT, Claude, GitHub Copilot CLI, Gemini and other MCP clients. It uses OAuth 2.1 and each person's existing Jira permissions. Connecting is free, but Atlassian says each MCP call consumes Rovo credits from your shared pool, even though you also pay for the assistant.

### What changes on 3 December 2026?

Atlassian starts billing for Rovo usage beyond the included credits, and extra usage is enabled by default. The list price is $0.01 per credit, or $10 per 1,000. Billing admins can see usage by app and user under Insights, Platform usage, Rovo credits, and can set spending limits or switch extra usage off before the date.

### Should AI set issue priority automatically?

Not at first. Let the model suggest severity against your written definitions and record its suggestion in a separate field, then have a person apply it. After a few hundred issues you can compare suggestions with decisions and see where it is reliable. Even then, keep people deciding anything that pages an on-call engineer or changes a customer commitment.


## Sources

- [Rovo usage limits](https://support.atlassian.com/rovo/docs/rovo-usage-limits/)
- [Rovo licensing and credits](https://www.atlassian.com/licensing/rovo)
- [Get started with the Atlassian Rovo MCP server](https://support.atlassian.com/atlassian-ai-gateway/docs/get-started-with-the-atlassian-remote-mcp-server/)
- [Jira Cloud platform REST API v3](https://developer.atlassian.com/cloud/jira/platform/rest/v3/intro/)
- [Rate limiting](https://developer.atlassian.com/cloud/jira/platform/rate-limiting/)
- [Webhooks](https://developer.atlassian.com/cloud/jira/platform/webhooks/)
- [OWASP Top 10 for LLM Applications 2026](https://genai.owasp.org/resource/owasp-genai-llm-top-10-2026/)
- [Pricing](https://platform.claude.com/docs/en/about-claude/pricing)

Published by Levy Street (https://levystreet.com), Wellington, New Zealand.