Issue: Authenticated MCP initialize to the Atlassian remote MCP server times out consistently.

Question

Behaviour:

* Unauthenticated POST to https://mcp.atlassian.com/v1/mcp returns 401 in ~0.1s.
* OAuth completes successfully — a valid access token (expires_in ~28700s) and refresh token are issued.
* With that token, initialize times out after 25–30s. Reproduced consistently across ~10 attempts.
* Same behaviour on both https://mcp.atlassian.com/v1/mcp and https://mcp.atlassian.com/v2/mcp.

Ruled out on our side: client binary (a public no-auth MCP server over the same POST+SSE transport connects in 0.23s from the same machine and process), local security software, proxy, DNS, TLS interception, OAuth scopes, callback port.

Ciro · Answer

Hi! One thing missing from the thread: Atlassian logged an incident for exactly this window. The Rovo status page shows "Degraded performance of Rovo MCP" on Oct 21, 15:30–18:25 UTC, and the 16:35 update says affected users may see "timeout issues". Your post went up at 15:55, and a clean 401, then valid tokens, then a hanging initialize looks a lot more like a struggling server than an auth or admin-settings problem.

It's marked resolved and the MCP component shows operational now, so I'd rerun the same handshake before changing anything on your side. If initialize still times out today, that's a new problem worth a ticket, and your write-up is already most of it.

Artyom Karpets _ArtUp Labs_ · Answer

Hi Daisy, a few notes that may save you time. I cannot see the server side, so these are checks, not a fix.

Your test is already clean: 401 unauthenticated, tokens issued, then a timeout on initialize. That points past OAuth and past your client. Three other people reported the same pattern in this forum within the last day (Claude desktop and Cowork connectors), so it may not be specific to you.

Endpoint: Atlassian's setup page ("Getting started with the Atlassian Rovo MCP server") lists mcp.atlassian.com/v2/mcp and says v1 usage moves to v2 tools on 1 March 2027. Since you already tried both URLs, the remaining variable is the account and its organization: an org admin decides which AI tools and domains may connect over OAuth and which permissions (read, write, search) the server gets, under Atlassian Administration > Connectors > Atlassian MCP server ("Control Atlassian MCP server settings"). Worth asking your admin whether your client's domain is allowed there.

If you have a second Atlassian site or account, repeating the same handshake there tells you whether the problem follows the account. If you open a support ticket, include the response headers and the time (UTC) of one failed call, since a request ID lets them trace it.

status.atlassian.com showed no incident when I looked this morning (22 Oct, 04:09 UTC).

Drafted with an AI assistant and reviewed by me