ChatGPT Atlassian Rovo MCP connector works briefly after reconnect, then becomes unavailable

Question

I’m having a persistent issue with the Atlassian Rovo MCP connector in ChatGPT.

The integration was working normally a few days ago and I was able to read and create Jira issues from ChatGPT. It now repeatedly becomes unavailable shortly after reconnecting.

Environment

* ChatGPT Plus personal account

* Normal ChatGPT project chat, not a custom GPT with Actions

* Atlassian Rovo plugin/connector

* Jira Cloud site: jamiembowers.atlassian.net

* Jira project: PWB

* Reproduced on both my work network and my home network

What happens

* Atlassian Rovo is installed and connected in ChatGPT.

* At one point, Jira requests failed with:

403 — The app is not installed on this instance.

* I disconnected and reconnected Atlassian Rovo.

* After reconnecting, ChatGPT successfully accessed Jira again and was able to read issues from the PWB project.

* Within minutes, later requests failed because the Atlassian Rovo connector/tool became unavailable before the Jira request could complete.

* This has happened even on simple read operations such as listing accessible Atlassian resources.

* I changed the ChatGPT plugin permission to Allow all actions/full access, but the issue continued.

* I reproduced the same behavior from both my work network and my home network, so it does not appear to be specific to a corporate firewall or proxy.

Recent reproduction timeline

All times are US Eastern Time.

* Aug 12, 2026 around 9:52–9:55 AM EDT: Reconnected Atlassian Rovo. Jira access temporarily worked and ChatGPT successfully read my PWB project.

* Shortly afterward, the connector became unavailable again during later Jira operations.

* Aug 12 around 11:18 AM EDT: Retried after granting full access. The connector again became unavailable.

* Aug 12 around 5:37 PM EDT: Retried from my home network. The connector again became unavailable while attempting a basic Atlassian resource lookup.

Expected behavior

Once OAuth authorization succeeds, the ChatGPT connector should remain available and continue to access the authorized Atlassian site until the user disconnects it or authorization is revoked.

Actual behavior

Reauthentication can temporarily restore access, but the connection becomes unavailable again shortly afterward.

OpenAI Support has also been given these details and has escalated the issue to a support specialist.

Could someone from the Atlassian Rovo MCP team help determine whether this could involve:

* OAuth token/session refresh

* tenant/site installation mapping

* MCP session persistence

* connector availability after initial authentication

* the ChatGPT client integration with the Atlassian Rovo MCP Server

I can provide screenshots of the ChatGPT plugin settings, timestamps, and additional reproduction details if useful.

Jeremiah Bowers · Answer

No problem @Tomislav Tobijas. I hope you enjoyed your vacation. OpenAI has escalated this issue and told me they would be in contact with me in a few days. If they are unable to resolve it, I will look into your suggestion for Atlassian Support

Tomislav Tobijas · Answer

Kudos for sharing all of these details @Jeremiah Bowers ! (and apologies for the delay; I got back home from vacation)

I'm not really sure what else you could check 🫤 I am using the same configuration when it comes to ChatGPT connector, the same AI plan, etc. and I haven't experienced any issues so far (and I'm using it daily).

If OpenAI isn't able to resolve this, I'd recommend reaching out to Atlassian Support (as stated above) as they can probably share some additional insights. 👀

Jeremiah Bowers · Answer

Thanks, @Tomislav Tobijas . I worked through the checks you suggested and wanted to report back with the results.

Things we had already tested

Before your reply, we had already confirmed:

* The issue reproduces on multiple devices, including my phone, work PC, and home PC.

* It reproduces on both my work network and home network, so it does not appear to be specific to a corporate firewall or proxy.

* Disconnecting and reconnecting Atlassian Rovo in ChatGPT can briefly restore access, but the connector becomes unavailable again shortly afterward.

* The problem has occurred during both read and write operations.

* At one point the runtime returned:
403 — The app is not installed on this instance.

* After a reconnect, ChatGPT was briefly able to access Jira again and successfully read issues from my Jira project, but later calls failed because the Atlassian Rovo tool itself became unavailable/disabled.

* I also tested with the ChatGPT plugin permission set to Allow all actions/full access, and the problem still occurred.

System clock check

I checked the Windows time service using:

w32tm /query /status

The result looked healthy:

* Leap Indicator: 0 (no warning)

* Source: time.windows.com

* Last Successful Sync Time: 8/13/2026 6:18:03 PM

* NTP synchronization was active

So system clock drift does not appear to be the issue.

Atlassian Rovo MCP permissions

I checked the Atlassian Admin Hub under:

Rovo → Rovo MCP Server → Permissions

All permissions were enabled:

* Read: 8/8

* Write: 8/8

* Search: 3/3

So the Atlassian-side MCP permissions also look correct.

Clean OAuth reset test

I then followed your stale-grant recommendation:

* Went to id.atlassian.com/manage-profile/apps

* Found Atlassian MCP

* Clicked Remove access

* Waited the full 10 minutes

* Disconnected Atlassian Rovo from ChatGPT

* Reconnected Atlassian Rovo and completed a fresh authorization

Immediately after that clean reset, we tested the connection again.

ChatGPT could still see that the Atlassian Rovo tool existed, but on the first actual call to list accessible Atlassian resources, the Atlassian Rovo connector/tool became unavailable again before returning the Jira site information.

So the issue still reproduces after:

* multiple devices

* multiple networks

* correct system time

* correct Atlassian Read/Write/Search permissions

* ChatGPT full-access permissions

* revoking the Atlassian MCP OAuth grant

* waiting 10 minutes

* disconnecting ChatGPT

* performing a completely fresh reconnect

At this point it looks less like a local configuration issue and more like something involving connector/session persistence, MCP runtime availability, OAuth/session handling, or the ChatGPT client integration itself.

OpenAI Support has also escalated the issue on their side.

If there are any additional Atlassian-side logs, request IDs, MCP diagnostics, or tenant/session information I can collect, let me know where to find them and I’m happy to provide them.