@Tom Polten _RCS_
Welcome to the Community, and good news: based on everything you have verified, nothing is misconfigured on your side. As of today, Rovo not appearing in the Agent Builder app list is expected behaviour, not a fault.
Here is the distinction that explains it. Atlassian's official ChatGPT integration is the Rovo MCP connector for ChatGPT, and Atlassian's supported-clients documentation lists "OpenAI ChatGPT" as the client. Nothing in Atlassian's documentation covers the Agent Builder app catalog, which is a separate surface, and which apps appear there is determined on OpenAI's side, not by anything in your Atlassian configuration. So your workspace approval and working account connection can all be correct, and Rovo still will not show in that list today.
The path worth trying now: if your ChatGPT plan allows adding a custom MCP server as a tool in Agent Builder, you can connect the Rovo MCP Server directly rather than waiting for the catalog. Two details matter if you go this route:
- Use the current endpoint:
https://mcp.atlassian.com/v1/mcp/authv2. Atlassian retired the older https://mcp.atlassian.com/v1/sse endpoint after 30 June 2026, so any setup instructions or existing configurations referencing the SSE endpoint stopped being supported days before you posted. Worth re-checking anything you configured earlier against that date.
- Authentication: OAuth 2.1 through the browser flow, or an API token if your Atlassian org admin has left API token authentication enabled. If token auth is disabled at the org level, OAuth is the only route, and that is an admin.atlassian.com setting, not a ChatGPT one.
And one set of Atlassian-side settings to have your org admin check, because they gate whether any connection works once added. In Atlassian Administration there is an Atlassian Rovo MCP server page (I verified this on a live org today) with three controls that matter here:
- Domains: every AI tool is associated with a domain, and access is domain-gated. There is an "Allow Atlassian supported domains" toggle for the tools Atlassian partners with, plus a list for adding your own trusted domains. On an org where that toggle is off and no domains are added, no AI tool can connect at all, whatever you configure on the ChatGPT side. This is the first thing to check.
- Permissions: access can be allowed or blocked by intent (Read, Write, Search), so a connection can exist but still refuse the action your agent attempts.
- Authentication: the API token option for non-interactive access lives here; if it is disabled, OAuth 2.1 is the only route.
- And a fourth tab worth noticing: A2A, marked Beta. Per the tab itself, this allows third-party AI agents to communicate directly with Rovo using the Agent-to-Agent protocol: when the single Allow A2A toggle is enabled, external agents can discover Rovo's capabilities and collaborate on tasks across Atlassian apps. It uses OAuth 2.0 only, no API tokens, and it is off by default. Not your immediate problem, but if you are building agents against Atlassian, this tab is where the next chapter starts.
And note that MCP clients act with the connecting user's existing Jira and Confluence permissions, so a least-privilege account for agent use is worth considering from day one.
On the bigger question of when or whether the official Rovo connector surfaces inside the Agent Builder catalog itself: that is an OpenAI-side decision, and neither company has published anything on it that I can find. The Atlassian Rovo MCP Server community space here is where the team building it answers questions, so it is worth raising there too; if I see an announcement, I will come back and update this answer.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.