Atlassian MCP: Confluence and Jira in Claude Code and Cursor

Updated 9 min read

“Atlassian MCP” usually means one of two servers. Atlassian’s own, now called the Atlassian MCP server, is hosted at https://mcp.atlassian.com/v2/mcp. It signs each person in with OAuth 2.1, works only with Atlassian Cloud and draws on your organization’s Rovo credits. The open-source mcp-atlassian (MIT, 5,988 GitHub stars) runs on your machine or in Docker with an API token, and it is the one that works with Confluence and Jira Server and Data Center. A Cloud team that wants admin controls and an audit log should start with Atlassian’s.

We have not run either server. Everything below, setup steps included, comes from Atlassian’s support docs, the two GitHub repositories, the mcp-atlassian docs and Anthropic’s Claude Code MCP docs as they stood on 12 October 2026.

What is the Atlassian MCP server called now?

Atlassian’s support docs title it the Atlassian MCP server. You will also see Atlassian Rovo MCP Server (in the docs navigation and Atlassian Administration) and Atlassian Remote MCP Server (in older URLs). All three names mean the same hosted server.

The current version is v2, at https://mcp.atlassian.com/v2/mcp. It covers Jira, Confluence, Jira Service Management, Bitbucket Cloud, Compass, Loom, and platform data such as Goals, Projects, Teams, Focus, Talent and the Teamwork Graph. Atlassian’s product page counts 215+ tools across 10+ apps. Atlassian stopped supporting the legacy SSE endpoint, https://mcp.atlassian.com/v1/sse, after 30 June 2026.

The atlassian/atlassian-mcp-server repository (Apache-2.0, last pushed 10 October 2026) holds plugin packages, skills and docs; the server itself runs on Atlassian Cloud.

Which clients does the Atlassian MCP server support?

It works with any MCP client that can reach a remote server. The docs give setup steps for VS Code with GitHub Copilot, Cursor, Claude Code, Claude Desktop, Codex and Codex Desktop, and Windsurf, and list ChatGPT, Claude, Docker, GitHub Copilot CLI, Google Gemini CLI, Kiro and Amazon Quick Suite as supported. A local client that cannot reach a remote server can go through the mcp-remote proxy, which needs Node.js 18 or later. In Claude on the web, Atlassian is in the connector directory (see Claude connectors).

Official server or mcp-atlassian: which should your team use?

Atlassian MCP server mcp-atlassian
Made by Atlassian Community; “Not an official Atlassian product”
License Apache-2.0 (repo); server hosted by Atlassian MIT
Runs Atlassian Cloud Your machine (uvx, pip), Docker, or your own HTTP service
Products Jira, Confluence, JSM, Bitbucket, Compass, Loom and more Jira and Confluence
Deployments Cloud only Cloud, Server and Data Center (Confluence 6.0+, Jira 8.14+)
Sign-in OAuth 2.1 per user; API token if an admin enables it API token, Personal Access Token, or OAuth 2.0
Tools 215+, most discovered on demand 98
Cost Calls consume Rovo credits Free software; you host it
GitHub, 12 Oct 2026 1,089 stars 5,988 stars, 1,392 forks, v0.23.1 (19 Aug 2026)

Pick Atlassian’s server if you are on Cloud and want each developer to sign in as themselves, with admins able to limit, log and revoke access centrally. Pick mcp-atlassian if you run Server or Data Center, need to keep traffic on your own network, or want to filter down to a few tools and spaces in config. The mcp-atlassian docs we read say nothing about Rovo credits.

How do I add the Atlassian MCP server to Claude Code?

From Atlassian’s docs:

claude mcp add --transport http atlassian https://mcp.atlassian.com/v2/mcp

Then open a session and run /mcp to sign in through your browser. To share the server with the whole team, add --scope project. That writes the URL, and no secret, to .mcp.json in the repository; each developer then signs in with their own Atlassian account. The Claude Code MCP guide covers scopes and .mcp.json.

claude -p has no /mcp panel, so it cannot run an OAuth sign-in itself: sign in first from an interactive session, or, for CI and other headless runs, use the API token route described under admin controls below.

For Codex, run codex mcp add atlassian --url https://mcp.atlassian.com/v2/mcp (see the Codex MCP guide).

How do I add it to Cursor?

Atlassian’s docs send Cursor users to the Atlassian plugin on the Cursor Marketplace: open it and select Add to Cursor. The support page and the repository README also have an Add to Cursor quick-install link. Atlassian does not document a hand-written Cursor config; the Cursor MCP guide covers Cursor’s MCP settings in general. If you route MCP through a gateway that needs every tool listed up front, Atlassian documents https://mcp.atlassian.com/v2/mcp?tools=all.

How do I run mcp-atlassian with uvx or Docker?

Create an API token at https://id.atlassian.com/manage-profile/security/api-tokens for Cloud, or a Personal Access Token on Server and Data Center. The README’s configuration for Cursor or Claude Desktop runs the server with uvx (shown here for Confluence only; add the matching JIRA_ variables for Jira):

{
  "mcpServers": {
    "mcp-atlassian": {
      "command": "uvx",
      "args": ["mcp-atlassian"],
      "env": {
        "CONFLUENCE_URL": "https://your-company.atlassian.net/wiki",
        "CONFLUENCE_USERNAME": "your.email@company.com",
        "CONFLUENCE_API_TOKEN": "your_api_token"
      }
    }
  }
}

The same in Claude Code:

claude mcp add \
  -e CONFLUENCE_URL=https://your-company.atlassian.net/wiki \
  -e CONFLUENCE_USERNAME=you@company.com \
  -e CONFLUENCE_API_TOKEN=your_api_token \
  --transport stdio mcp-atlassian -- uvx mcp-atlassian

The Docker image from the mcp-atlassian docs:

docker run -i --rm \
  -e CONFLUENCE_URL=https://your-company.atlassian.net/wiki \
  -e CONFLUENCE_USERNAME=you@company.com \
  -e CONFLUENCE_API_TOKEN=your_api_token \
  ghcr.io/sooperset/mcp-atlassian:latest

In an MCP client, that goes in as "command": "docker" with the rest as args. On Server and Data Center, replace the username and API token with CONFLUENCE_PERSONAL_TOKEN and JIRA_PERSONAL_TOKEN.

The settings a team lead will want:

  • READ_ONLY_MODE=true turns off every write tool.
  • CONFLUENCE_SPACES_FILTER=DEV,TEAM,DOC limits the agent to those spaces.
  • ENABLED_TOOLS=confluence_search,jira_get_issue allows only the tools you name.
  • TOOLSETS=default loads the core toolsets only. Left unset, it currently loads all of them.

It can also run as one shared service with --transport streamable-http, where each client sends its own token. The token carries its owner’s access, so keep it out of committed files.

How do agents search, read and write Confluence pages?

Both servers act as the signed-in user, so an agent sees and edits only the spaces that person can.

On Atlassian’s server, Confluence tools sit in three permission groups: read_confluence, write_confluence and search_confluence. Search across products runs through Rovo search tools (the docs name search and searchAtlassian), which consume Rovo credits. You ask in plain language, such as “Summarize the Q2 planning page.” Only a few primary tools load when the client connects; the agent discovers the rest as needed, which saves context.

On mcp-atlassian, the core confluence_pages toolset has 14 tools. The ones agents use most:

  • confluence_search takes CQL, such as space = DEV AND type = page AND lastModified < '2024-01-01'.
  • confluence_get_page reads a page by ID, with metadata if asked.
  • confluence_create_page takes a space key, a title, markdown content and an optional parent page ID.
  • confluence_update_page and confluence_update_page_section edit an existing page.
  • confluence_get_page_history and confluence_get_page_diff show what changed.

Every write is a real Confluence edit under that person’s name. For a first rollout, READ_ONLY_MODE=true on mcp-atlassian, or leaving write_confluence off on Atlassian’s side, lets agents read and search before they write.

What can Atlassian admins control?

From Atlassian’s docs and product page, for the official server:

  • Admins grant or revoke tools by permission group. delete_jira and manage_jira are off until an admin turns them on.
  • Data policies allow or block AI access by site, space or classification, and sign-in can go through your own identity provider and SSO.
  • The Rovo MCP server settings page in Atlassian Administration decides which AI tools and domains may connect over OAuth. Atlassian-supported domains are allowed by default.
  • IP allowlists you set for Atlassian Cloud apply to MCP calls too.
  • API token sign-in is off until an organization admin enables it under Atlassian Administration, then Rovo, then Rovo MCP server, then Authentication. After that, a client can use a personal API token (Basic auth) or a service account API key (Bearer). Jira Service Management tools only work with a token.
  • Every tool call lands in the organization audit log, under Insights, then Audit log (filter for Rovo MCP User Actions).
  • Admins can revoke the app for the organization, and each user can revoke their own authorization.

mcp-atlassian has none of this centrally: its limits are whatever each configuration sets.

How is Rovo different from the Atlassian MCP server?

Rovo is Atlassian’s AI inside its own apps: Rovo Chat, Studio, Agents and search, built on the Teamwork Graph. The MCP server lets AI tools outside Atlassian, such as Claude, ChatGPT, Cursor, Codex and Copilot, use the same data under the same permissions.

They share one budget: Atlassian’s docs say MCP calls consume Rovo credits from the same organization-wide pool as Rovo Chat, Studio and Agents, scaled by how much data and reasoning a call needs. Per-plan allowances are on Atlassian’s Rovo usage limits page, which we did not check. If your developers work in Claude Code or Cursor, the MCP server brings Jira and Confluence to them; in Confluence’s own interface, Rovo is already there.

Where does a wiki your agents keep fit next to Confluence?

Confluence holds what people write for people: specs, runbooks, decisions with sign-off. Agents also pick things up as they work, like which workaround fixed a flaky test or what another developer’s agent found yesterday. Writing all of that into Confluence mixes session notes into spaces people curate.

A separate wiki that the agents keep can hold that working knowledge, with every agent on the team reading and writing the same pages. Dexio is a hosted wiki of that kind. Claude Code, Codex, Cursor, Claude, ChatGPT and other agents read, search, write and link its markdown pages over MCP, and people see what their agents know as a page graph. Each change is stored with the agent and person behind it. Several agents we run share one Dexio wiki; on 12 October 2026 it held 107 pages and 565 links between them. In another workspace we run, agents wrote 189 pages on its first day (26 September 2026).

Use Atlassian’s server for specs, tickets and Jira updates, and Dexio for what the agents learn. Confluence is the better place when knowledge needs space permissions, data classification and an audit log, or when people edit it there every day. Notion teams face the same split (see Notion MCP), and the company wiki guide covers where team knowledge should live.

Dexio is free for one person; Team is $10 and Business $20 a member a month, and agents never count as members. It is open source under AGPL-3.0 if you would rather host it yourself. The server is https://app.dexio.wiki/mcp with an API key in a header, so it works in headless runs too. To connect an agent, send it:

Connect yourself to my Dexio wiki. The steps are at https://dexio.wiki/agents.md: read the whole file, not a summary, and follow them.

FAQ

Is the Atlassian MCP server free?

Atlassian’s product page says “Start for free. No credit card required.” Calls still consume Rovo credits from your organization’s pool, with allowances set by plan. mcp-atlassian is free under the MIT license, and you run it yourself.

Does the Atlassian MCP server work with Confluence Data Center?

No. It needs an Atlassian Cloud site. For Server and Data Center, mcp-atlassian supports Confluence 6.0 and later and Jira 8.14 and later, with a Personal Access Token.

Do v1 connections still work?

The v1 /mcp endpoints are still supported and switch to v2 tools automatically; the support page dates that switch 1 March 2027. Clients that then fail to sign in may need to clear cached client IDs or .well-known credentials. New setups should use https://mcp.atlassian.com/v2/mcp.

How do I add Jira MCP to VS Code?

In the Extensions view, search @mcp Atlassian, select the Atlassian MCP server and choose Install. The same server covers Jira and Confluence.

Sources

Ask a question

Ask anything about Dexio.

About
We reply by email.