"GPT" and "Atlassian Rovo" connector by GPT APPs

[Vladislav Popov]

Hello Atlassian Support team!

I’m currently using Atlassian Rovo in Jira and Confluence and would like to clarify the intended access model and recent changes related to external AI integrations.

Context:
- Atlassian Rovo is enabled and actively used inside Jira and Confluence UI.
- I also connected the Rovo connector to ChatGPT, expecting controlled interaction with Jira/Confluence data.
- As of now, ChatGPT does not have any read or management access:
- no page listing or reading (Confluence)
- no issue listing or reading (Jira)
- no Rovo search or query methods available

Important clarification:
- Until approximately **16.12.2025**, ChatGPT was able to interact with Jira via external Jira APIs:
- create and update issues,
- add comments,
- modify fields,
- link issues,
using authenticated API-based integration.
- At the moment, this capability is no longer available in the same way, and Jira operations from ChatGPT are fully unavailable.

Questions:
1) Was the removal or restriction of external Jira API operations from ChatGPT an intentional product or security change?
2) Is Atlassian Rovo designed to operate strictly inside Atlassian UI without exposing data or actions to external AI tools?
3) Is there any supported or recommended way to:
- query Jira / Confluence data,
- or manage issues,
using external AI systems (e.g. ChatGPT)?
4) Are API-based or connector-based integrations planned or officially supported on the roadmap?
5) Are there recommended enterprise-safe patterns for combining Rovo with external AI tools (manual sync, export workflows, approved APIs)?

Goal:
I want to design a compliant, secure, and future-proof workflow that combines Atlassian knowledge (Jira / Confluence) with AI-assisted analysis, without violating security or product constraints.

Thank you in advance for the clarification.
Best regards, Vladislav

[Vladislav Popov]

In addition, I would like to note the following points:

On this page: https://support.atlassian.com/atlassian-rovo-mcp-server/docs/setting-up-clients/

And in the APPs tab in GPT(Screenshot),


there is a mention that all of this should work, as this is the essence of this connector for GPT. But now, when I ask GPT to perform any operation, even reading information from Jira, it tells me that when it makes a request

api_tool.list_resources

it gets an empty list of methods:

{ “finite”: true }

Thus, GPT cannot perform any of the operations stated in the description of the Atlassian Rovo connector.

[Vladislav Popov]

Today: 19.12.2025

---

What happens when I run

“Check my project with key KAN and analyze the current board using Atlassian Rovo”

I attempted to execute this request right now using the Atlassian Rovo (MCP) connector.
The request did not reach Jira, because the Atlassian MCP gateway is currently unavailable in this session.

---

Step-by-step: actual execution flow

1) Tool discovery

The Atlassian Rovo connector is present and exposes Jira/Confluence operations (read/write capabilities are advertised).

This means the integration is configured, not missing.

---

2) First mandatory MCP call

Before any Jira request can be made, the integration must resolve the Atlassian cloudId.

This is done via:

getAccessibleAtlassianResources()

This call is required to:

• identify which Atlassian Cloud sites are accessible

• obtain cloudId values used by all Jira and Confluence APIs

---

3) Failure point (hard stop)

The call failed with a transport-level error:

• HTTP 424

• Underlying error: 524 Unknown

• Endpoint: https://mcp.atlassian.com/v1/sse

Plain English:

The MCP server did not respond.
This is not a permissions issue and not a project-level restriction.
The request never reached Jira.

Because of this failure, no Jira APIs can be invoked.

---

What would happen if MCP were available

If the MCP endpoint responded correctly, the execution pipeline would look like this:

1) Resolve project

getVisibleJiraProjects(
  cloudId,
  searchString = "KAN",
  action = "browse"
)

Confirms project visibility and retrieves metadata.

---

2) Reconstruct “current board” state

(Board access is typically inferred via issue queries.)

Example JQL queries:

project = KAN
AND statusCategory != Done
ORDER BY updated DESC

or (Scrum projects):

project = KAN
AND sprint IN openSprints()

Executed via:

searchJiraIssuesUsingJql(
  cloudId,
  jql,
  fields = [
    status,
    assignee,
    priority,
    issuetype,
    labels,
    sprint,
    created,
    updated
  ]
)

---

3) Analysis layer

From the retrieved data, the analysis would include:

• WIP distribution by status and assignee

• Aging issues (stale tickets)

• Blockers and dependencies (issue links)

• Priority imbalance

• Actionable recommendations (what to close, split, or reassign)

---

Current conclusion

At the moment, KAN cannot be analyzed via Rovo, because:

• The Atlassian MCP SSE endpoint is unreachable (524)

• The failure occurs before authentication or authorization

• This is a platform connectivity issue, not a Jira permission problem

In short:

The keys are valid, but the door is not opening.