I get this error message when I tried to link to a valid confluence or jira url.
You'll need to check in with your Atlassian Org Admin - unless you happen to be one - and check the configuration of the Rovo MCP settings.
https://support.atlassian.com/atlassian-ai-gateway/docs/use-atlassian-rovo-mcp-server/
The area of interest you may want to check is the allow list - if claude's domain isn't listed - it will be blocked.
The other thing is ensuring Confluence is permitted to talk to Rovo and isn't on the block list.
Other area to check would be Rovo's overall global policies - like talking to external URL's (but in this context its MCP - not Rovo itself)
Hopefully this helps - if not at least give you a good start.
The allow list is worth checking, but there is a more likely cause worth ruling out first, and it is one Atlassian states plainly in its own tracker: Rovo accesses data within its connected Atlassian site. If the Confluence or Jira URL you pasted lives on a different site from the one the MCP connection is bound to, it will fail however valid the URL is, and no allowlist entry changes that.
The documented route for a second site is Atlassian Admin, then Apps, Sites, pick the site, Connected Apps, Add external MCP server, with the MCP URL set to https://mcp.atlassian.com/v2/mcp. The catch is that this path is currently broken: AX-2224, "External MCP OAuth consent flow returns HTTP 431 when connecting a second Atlassian site", is at Needs Triage and was last updated on 1 October. So if your error is an HTTP 431, you have found that one rather than a misconfiguration on your side.
On the error message itself being unhelpful, that is also tracked. ECO-1630 asks the Rovo MCP Server to return an actionable error when a token or request is missing orgId or a valid cloudId, instead of the generic "We are having trouble completing this action". Worth knowing, because the wording you are seeing may be hiding a missing context value rather than a permission problem.
One more gate that catches people: calls made through Rovo MCP consume Rovo credits from the organisation pool, so an exhausted pool produces failures that look like access errors.
If you can post the exact error text, and say whether the URL is on the same site Rovo is connected to, that narrows it to one of these immediately.
I would first test with the canonical Confluence page URL (the site URL plus the page path), not a copied smart link or an embedded view. The Atlassian account used in the OAuth flow must have access to that site and page, and some organizations require an admin to approve the connector/scopes. In Claude, remove the stale Atlassian connection and authorize it again, then retry with a page you can open directly in the same browser account. Avoid pasting any access token into the thread; if consent still fails, the redacted error, site URL, UTC time, and whether other pages connect will distinguish permissions from an OAuth redirect issue.
It looks like you're new here. Sign in or register to get started.
