Atlassian Rovo MCP Deep Dive

You are three hours into implementing a feature when you realize the Confluence page linked in the Jira ticket has been updated since you last read it. The acceptance criteria changed. The API contract in the architecture doc was revised. You have been building against stale requirements, and now half your implementation needs reworking. If only your AI had read the current docs before you started coding.

The Atlassian Rovo MCP server gives your AI direct access to Jira issues, Confluence pages, and Compass components through a single remote MCP endpoint. The AI reads the current ticket, pulls in linked documentation, and works from the latest version of every requirement — no manual context-switching required.

What You’ll Walk Away With

• Full setup for the Atlassian Rovo MCP server with OAuth 2.1 authentication
• Prompts for ticket-driven development, documentation search, and cross-tool workflows
• Understanding of available tools: Jira issues, Confluence pages, Compass components
• Troubleshooting for OAuth flows, permission errors, and site admin requirements

How Rovo MCP Works

Unlike most MCP servers that run locally on your machine, the Atlassian MCP server is a remote server hosted by Atlassian at https://mcp.atlassian.com/v1/mcp. Authentication happens through OAuth 2.1 — the first time you connect, a browser window opens for you to authorize access to your Atlassian site. After authorization, the MCP client caches your credentials.

The server provides tools across three Atlassian products:

Product	Capabilities
Jira	Search issues with JQL, read issue details, create issues, update status, add comments, manage sprints
Confluence	Search pages, read page content, create pages, update existing pages
Compass	Browse components, read component metadata, understand service dependencies

Setup

Cursor

{
  "mcpServers": {
    "atlassian": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.atlassian.com/v1/mcp"]
    }
  }
}

When Cursor first connects, a browser tab opens for OAuth authorization. Select your Atlassian site, grant the requested permissions, and the connection completes automatically.

Claude Code

{
  "mcpServers": {
    "atlassian": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.atlassian.com/v1/mcp"]
    }
  }
}

Add this to .claude/settings.json or use the CLI:

claude mcp add atlassian -- npx -y mcp-remote https://mcp.atlassian.com/v1/mcp

The OAuth flow opens in your default browser on first use.

Codex

[mcp.atlassian]
transport = "http"
url = "https://mcp.atlassian.com/v1/mcp"

Note

The first user on your Atlassian site must have access to both Jira and Confluence for the initial OAuth consent. If you see “Your site admin must authorize this app,” ask your Atlassian admin to complete the first authorization.

Jira Workflows

Reading and Understanding Tickets

The most immediate value is letting the AI read the full context of a ticket before you start coding. No more switching to the browser to re-read requirements.

Tip

Copy-paste prompt for ticket-driven implementation:

“Search Jira for ticket PLATFORM-1234. Read its description, acceptance criteria, and any linked issues. Also fetch any Confluence pages linked in the ticket. Based on all of this context, outline the implementation approach and identify any ambiguities I should clarify with the PM before coding.”

Creating and Updating Tickets

The AI can also write back to Jira — creating bug reports from code analysis, updating ticket status as you work, or adding implementation notes as comments.

Tip

Copy-paste prompt for automated bug reporting:

“I just found a race condition in src/lib/auth/session.ts where two concurrent requests can invalidate each other’s sessions. Create a Jira bug ticket in the AUTH project with priority High. Include the file path, line numbers, a description of the race condition, steps to reproduce, and a suggested fix. Assign it to me.”

Sprint Management

Tip

Copy-paste prompt for sprint analysis:

“Search Jira for all issues in the current sprint of the PLATFORM project. Group them by status (To Do, In Progress, In Review, Done). For any ticket that has been In Progress for more than 3 days, read its description and comments to identify potential blockers. Summarize the sprint health.”

Confluence Workflows

Searching Documentation

Confluence search through MCP is faster than using the Confluence web UI because the AI processes results immediately rather than requiring you to click through and read each page.

Tip

Copy-paste prompt for architecture research:

“Search Confluence for pages related to our payment processing architecture. Read the top 3 most relevant results. Summarize the current architecture, identify any recent changes, and note any documented decisions about error handling in the payment flow.”

Creating Documentation

The AI can also create Confluence pages — useful for generating ADRs, runbooks, or technical specs from code analysis.

Tip

Copy-paste prompt for documentation generation:

“Analyze the database migration files in src/db/migrations/ from the last month. Create a Confluence page in the Engineering space titled ‘Database Schema Changes - February 2026’ that documents each migration: what changed, why (based on commit messages and linked tickets), and any breaking changes for downstream services.”

Compass Integration

If your organization uses Atlassian Compass for service catalog management, the MCP server can read component metadata, ownership information, and dependency maps.

This is particularly useful when working on a microservice and needing to understand what other services depend on it, who owns them, and what their SLAs are.

Combining Atlassian MCP with Other Servers

The strongest workflows combine Atlassian MCP with version control:

1. Read the ticket. The AI fetches the Jira issue, its acceptance criteria, and any linked Confluence docs.
2. Create the branch. Using GitHub MCP, the AI creates a branch named after the ticket ID (e.g., feature/PLATFORM-1234).
3. Implement the change. The AI writes code grounded in the ticket’s actual requirements — not a paraphrased summary.
4. Open a PR. Using GitHub MCP, the AI creates a pull request that references the Jira ticket in the description.
5. Update the ticket. The AI moves the Jira issue to “In Review” and adds a comment with the PR link.

When This Breaks

OAuth flow never completes. The MCP remote proxy redirects to localhost for the OAuth callback. If you are on a VPN or corporate proxy that blocks localhost callbacks, the flow will hang indefinitely. Disconnect from the VPN temporarily, complete the OAuth flow, then reconnect. The cached credentials persist across VPN reconnections.

“Your site admin must authorize this app” error. This appears when no admin on your Atlassian site has completed the initial OAuth consent. The first authorization must be done by a user with admin-level access to both Jira and Confluence. After that, regular users can authorize normally.

AI cannot find the ticket. The AI searches using JQL, which requires the project key, not the project name. “Find bugs in PLATFORM” works because PLATFORM is the project key. “Find bugs in Platform Engineering” fails because that is the project name.

Permission denied on updates. The MCP server respects your Atlassian permissions. If you cannot edit a ticket in the Jira web UI, the AI cannot edit it through MCP either. This is a feature, not a bug.

Stale OAuth tokens. If the connection worked yesterday but fails today, your OAuth token may have expired. Remove and re-add the MCP server to trigger a fresh OAuth flow.

What’s Next

Project Management MCP Broader project management integration including Linear.

Version Control MCP GitHub and GitLab integration to complete the ticket-to-PR loop.