What Atlassian’s MCP server is
Atlassian hosts one remote MCP server for its cloud products. You do not install anything: you add a URL to your AI tool and sign in.
| What | Detail, as of November 2026 |
|---|---|
| Endpoint | https://mcp.atlassian.com/v2/mcp (v1 moves to v2 on March 1, 2027) |
| Products | Jira, Jira Service Management, Confluence, Bitbucket, Projects, Goals, plus Loom |
| Clients named by Atlassian | ChatGPT, Claude, Claude Code, Claude Desktop, GitHub Copilot (VS Code and CLI), Cursor, Codex, Windsurf, Google Gemini, Amazon Quick Suite, and any app with MCP support |
| Sign-in | OAuth 2.1; API token optional, and admins can switch it off |
| Permissions | Each user’s own Jira and Confluence permissions |
| Cost | No separate price listed; some search and Teamwork Graph tools use Rovo credits |
Older guides call it the Atlassian Rovo MCP server. It is the same product; version 2 sits in what Atlassian calls the AI Gateway.
Setting it up for a team
- Admin first. An org admin decides which AI tools may connect. In the MCP server settings, the domain allowlist sets which tools and domains can connect with OAuth 2.1. If your organization uses an IP allowlist, MCP requests must also come from an allowed IP address, also when they use an API token.
- Set permissions per app. In Atlassian Administration go to Connectors > Atlassian MCP server > Permissions. Allow or block read, write and search, per app if you want (for example: Confluence read and search, Jira read only). Atlassian says these settings take precedence over Connected Apps, and changes apply right away.
- Add the server in each AI tool. In Claude or ChatGPT that is a custom connector with the URL; in VS Code, Cursor, Claude Code or Codex it goes in that tool’s MCP configuration. Per tool, see how to connect an MCP server to Claude, ChatGPT, Copilot and Grok.
- Each person signs in with their own Atlassian account through OAuth. The AI then acts as that person.
- Watch the audit log. Atlassian itself advises least privilege, reviewing high-impact changes and monitoring audit logs.
The common misunderstanding
Many teams expect that once Jira and Confluence are connected, the AI “knows the project”. What it gets is search and fetch: it can look up a ticket or a page when it thinks to. If your Confluence has three versions of the onboarding guide, the AI may find the old one. If the real decision was made in a Slack thread and never written down, the AI cannot find it at all. The MCP server makes your Atlassian content reachable; it does not make it correct or complete.
A second misunderstanding: “the AI can only read”. Not by default. Write tools exist, and an agent can create or edit issues and pages if the user’s permissions and the admin settings allow it. Decide that on purpose in step 2. For the general risks, read is MCP safe to use in a company.
What this means for a team
The good part: one server, one set of admin controls, and every MCP-capable tool in your team can reach the same Jira and Confluence. The developer in Cursor and the project lead in ChatGPT see the same tickets.
What it does not carry is what each person’s AI learned along the way: that a certain customer’s tickets always need a second check, that “done” in your team means deployed, that a template changed last month. That stays in each separate chat or tool. If you also connect GitHub, see the GitHub MCP server; for what an MCP server does in general, see what is an MCP server.
Where Bombos fits
The Atlassian MCP server makes Jira and Confluence reachable; Bombos adds a shared memory on top that every AI tool in the team uses, with team knowledge and rules the AI learned about how you work. It reads what already exists in documents, mail and repos and shows what each answer rests on. It runs in our cloud or on your own server, and one person can try it first with their own agents.
Related questions
Is the Atlassian MCP server the same as the Rovo MCP server?
Yes. Atlassian's documentation used the name Atlassian Rovo MCP server and now calls version 2 the Atlassian MCP server, part of the Atlassian AI Gateway.
Does the Atlassian MCP server cost extra?
Atlassian's getting-started page lists no separate price, but some search and Teamwork Graph tools, such as searchAtlassian, consume Rovo credits.
Can the AI see Jira projects I have no access to?
No. According to Atlassian, actions through the MCP server respect each user's existing access controls and permissions.
What happens to the v1 endpoint?
Atlassian notes that v1 usage will move to v2 on March 1, 2027. New setups should use https://mcp.atlassian.com/v2/mcp.
Can admins stop the AI from editing tickets?
Yes. In Atlassian Administration under Connectors > Atlassian MCP server > Permissions, org admins can allow or block read, write and search per app. Changes apply immediately.
