Until recently, wiring an AI to Atlassian meant running separate community servers for Jira and Confluence and juggling API tokens. The Atlassian Remote MCP server — built on Rovo — changes that: it's an official, hosted, OAuth-protected endpoint that reached general availability in early 2026 and exposes 72+ tools across Jira, Confluence and Compass from a single connection. Here's how to set it up.
Why the hosted server is the right default
Because Atlassian hosts and operates it, there's nothing to provision and no token to paste into a config file. You add the remote MCP URL as a connector, authorise once through OAuth, and your existing Atlassian permissions carry over automatically — the agent sees exactly what your account sees, no more. That single-sign-on path is the reason to prefer it over a self-hosted community server for anything beyond local experimentation: no long-lived API key in plaintext, and the same Jira and Confluence permissions you already enforce.
Connecting it
In a client that supports remote OAuth connectors (Claude, Cursor, VS Code), add the Atlassian Remote MCP endpoint as a URL-based connector and complete the OAuth flow in your browser. The client stores the resulting token, not your password. Once authorised, the full tool surface appears — creating and transitioning Jira issues, searching with JQL, reading and updating Confluence pages, and querying Compass components. If your client only supports local stdio servers, you can bridge to the remote endpoint with mcp-remote, but a native OAuth connector is cleaner. See MCP config file location for where connectors are stored.
Verify
Ask: "List the open issues assigned to me in the current sprint." Real issues back confirms the OAuth scope and your project access line up. Then try a Confluence read — "Summarise our team's onboarding page" — to confirm both products are reachable through the one server.
The SSE deadline you can't miss
Atlassian is retiring the old HTTP+SSE transport for Rovo MCP on 30 June 2026. If you configured the connector months ago against a /sse endpoint, it will stop responding on that date — update it to the current Streamable HTTP endpoint now. If a working Atlassian connector suddenly throws connection errors around that window, this is the first thing to check. The full background is in MCP SSE deprecated: migrate to Streamable HTTP.
Security
The remote server already does the hard part by enforcing your Atlassian permissions and using OAuth instead of static keys. Keep that posture: authorise from a least-privilege account where you can, review the scopes you grant, and keep a human checkpoint on write actions — transitioning issues and editing pages are real changes to shared team state. See MCP security best practices and enterprise managed MCP connectors.
Going further
Prefer self-hosting, or need only one product? See the standalone Jira MCP setup and Confluence MCP setup, and the Jira agent profile. Atlassian fits a delivery loop with Linear and GitHub; wire it into a product-manager loadout or browse the productivity category.