Forums

Articles
Create
cancel
Showing results for 
Search instead for 
Did you mean: 

Setting up ChatGPT (OpenAI) MCP Server with Atlassian Rovo (Atlassian AI)

Gaia Tamir Azran November 20, 2025

hi, im trying to Set up ChatGPT (OpenAI) MCP Server with Atlassian Rovo (Atlassian AI) by this doc - https://support.atlassian.com/atlassian-rovo-mcp-server/docs/setting-up-chatgpt/

for that, i want to create a service account in jira 'service account'. will it work? also, i need to choose the scopes after creating the service account after choosing 'OAuth 2.0'. in the doc they dont say which scopes i need for the gpt connection. and there are 438! scopes. please assist firstly, if this type of user will work for me, and secondly, what scoped to choose.

1 answer

1 vote
Peter_DevSamurai
Atlassian Partner
November 20, 2025

Hi @Gaia Tamir Azran , 

Your method for a dedicated service account is what Atlassian recommends for Chatgpt MCP integration. You just need to make sure that this service user has browse permission on the projects/spaces you want ChatGPT to access. You can try the steps below to setup the service account with OAuth 2.0:

  • Create the service account

    • admin.atlassian.com → Directory → Users → Invite user
    • Use an email you control (ex: your company's shared mailbox).
    • Add the user to the groups that give read access to the Jira projects and Confluence spaces you want indexed.

  • Register the OAuth app (once per org)

  • Scopes you actually need (these are the ones Atlassian and OpenAI use for ChatGPT ↔ Rovo MCP):, you only need like 6-7 below scopes, and won't be needing all 438.

    • offline_access : Required for refresh tokens

    • read:me : Identify the service user

    • read:jira-work : Read issues, projects, boards, filters
    • read:jira-user : Read user details (for @mentions etc.)
    • read:confluence-content.all : Read all Confluence pages/spaces (or restrict to specific spaces if you prefer granular scopes :) )
    • read:confluence-space : Space-level read (if you use granular instead of .all)

    • read:confluence-props : Read page properties
    • #Note: If you want to be more restrictive (which is recommended for production), replace read:confluence-content.all with the granular space scopes for only the spaces you want indexed (ex: read:confluence-content:SPACEKEY).

  • Install the ChatGPT plugin
    • In ChatGPT → Plugins → Develop your own → follow the wizard.
    • Paste your OAuth Client ID and Secret when asked.
    • The first connection will ask the service account to consent – log in as the service account (or pre-consent as site admin if possible).
    • After this =>  The MCP server will be able to pull Jira issues and Confluence pages into ChatGPT securely.

You can refer to the below links:


I hope everything works out well for you!

Best,

Peter

 

Gaia Tamir Azran November 23, 2025

hi @Peter_DevSamurai i tried but keep getting 'internal server error' i established an app in atlassian developer and gave all required scopes

BTW your chatgpt flow is the old one- now its apps&connectors not plugins, so the url should be https://chatgpt.com/connector_platform_oauth_redirect

i think i did everything as excepted but something is not working for me.

Like Peter_DevSamurai likes this
Peter_DevSamurai
Atlassian Partner
November 24, 2025

Hi @Gaia Tamir Azran , 

Thank you for the update and I'll note this down! The "internal server error" is likely tied to this mismatch or a scope/consent hiccup during the connector setup. 

As corrected, change the OAuth Redirect URI to: https://chatgpt.com/connector_platform_oauth_redirect; besides this everything should still remain. If the error persists, can you share the error payload or screenshot. What's your Chatgpt plan -> This may affect the connector limits.  

 

Gaia Tamir Azran November 24, 2025

our chatgpt plan is enterprise 

here is the screenshot i get. nothing else

image.pngimage.png

attached screenshot of the app i created in jira developer:

image.pngimage.png

Peter_DevSamurai
Atlassian Partner
November 24, 2025

Hi @Gaia Tamir Azran ,

ChatGPT enterprise requires workspace admin approval for full MCP. YOu can check:
1) Re-verify scopes:

  • Your scopes look good, can you help me ensure the below are included: offline_access (for refresh), read:jira-work, read:confluence-content.all (or granular like read:confluence-content:SPACEKEY), and read:me.
  • Note: Workspace admins must approve scopes in ChatGPT settings > Workspace > Connectors > Enable for your app (under "Custom connectors").


2) Re-Authorize the Connector:


3) Enterprise workspace config:

  • As admin: ChatGPT > Settings > Workspace > Developer/Connectors > Ensure "Custom MCP connectors" is enabled for your team (beta for full write actions).
  • MCP URL: Confirm https://mcp.atlassian.com/v1/sse in connector setup—Enterprise supports it natively. 


4) If Still 500: Which happens more common than you think

  • Clear cache/cookies for auth.atlassian.com and chatgpt.com. You can try incognito.
  • Regenerate API token on service account (id.atlassian.com > Security > Tokens > Classic scopes including offline_access).
  • Test token validity: curl -H "Authorization: Bearer TOKEN" https://api.atlassian.com/ex/jira/{cloudId}/rest/api/3/myself (get cloudId from admin.atlassian.com > Site settings > General).

The Enterprise flows are more strict on consent. 

You can refer to this doc:


Best, 

Peter

 

Suggest an answer

Log in or Sign up to answer
DEPLOYMENT TYPE
CLOUD
PRODUCT PLAN
PREMIUM
TAGS
AUG Leaders

Atlassian Community Events