Jira MCP Server: Atlassian's Official Setup Guide

The best Jira MCP server for most teams is the one Atlassian runs itself. It's a hosted endpoint at https://mcp.atlassian.com/v2/mcp, it signs you in with OAuth 2.1, and it can only do what your Jira account is already allowed to do. One command adds it to Claude Code.

The catch: it's for Atlassian Cloud. If your company runs Jira Data Center or Server, the community project mcp-atlassian is the usual answer.

Quick summary:

• Official server: Atlassian MCP Server (some Atlassian pages still call it the Atlassian Rovo MCP Server). Covers Jira, Confluence, Jira Service Management, Bitbucket, Loom, and more.
• Endpoint: https://mcp.atlassian.com/v2/mcp
• Claude Code: claude mcp add --transport http atlassian https://mcp.atlassian.com/v2/mcp, then /mcp to sign in
• Auth: OAuth 2.1 by default; API tokens only if your org admin turns them on
• Cost model: single-product lookups (one Jira work item or Confluence page) and write operations are free; enriched, cross-product and unified search/context calls consume Rovo credits from your organization's shared pool
• Data Center / Server: use sooperset/mcp-atlassian, MIT-licensed and not an official Atlassian product

Everything below is from Atlassian's and the project's own documentation, checked in October 2026.

What the official Atlassian MCP server is

Atlassian describes it as "a cloud-hosted Model Context Protocol (MCP) server that gives your AI tools secure, real-time access to your data across Jira, Confluence, Bitbucket, and other Atlassian apps." If MCP itself is new to you, start with what MCP is and come back.

Three facts shape how you'll use it:

It's Cloud only. Atlassian's client setup page lists "an Atlassian cloud site with Jira, Compass, and/or Confluence" as the first requirement. There's no Data Center option in its docs.

It respects your existing permissions. The getting-started guide says OAuth 2.1 "ensures all actions respect users' existing access controls and permissions." If you can't see a project in Jira, the agent can't either.

It's on v2 now. Older tutorials show /v1/ URLs. Atlassian says that on March 1, 2027, any existing v1 usage "will automatically start to expose and utilize v2 tools," and clients that can't handle the switch will need to clear cached credentials. New setups should use the v2 URL.

Set up the Jira MCP server in Claude Code

Atlassian's documented command:

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

Then open a Claude Code session and run /mcp. Pick atlassian, and Claude Code opens the Atlassian consent screen in your browser. Approve access, and you're connected.

By default this adds the server to the current project only. Add --scope user if you want Jira available in every project. Our Claude Code MCP configuration guide explains the scopes.

Prefer a plugin? Atlassian also publishes an Atlassian plugin for Claude Code that bundles the same MCP connection with skills like /spec-to-backlog and /capture-tasks-from-meeting-notes:

claude plugin install atlassian@claude-plugins-official

Set up the Atlassian connector in Claude.ai

On claude.ai and Claude Desktop, you don't paste a command. Atlassian is in the connectors directory as Atlassian MCP, made by Atlassian, pointing at the same v2 URL.

1. Go to Customize → Connectors (or click + in a chat, then Connectors → Manage connectors).
2. Click +, find Atlassian, and click Connect.
3. Sign in to Atlassian and approve access.

On Team and Enterprise plans, an Owner has to add the connector for the organization first, according to Claude's connector help article. Each person still authenticates with their own Atlassian account.

Admin requirements to know before you start

Most "it won't connect" problems are admin settings, not client bugs. Atlassian's admin controls page and its Anthropic partnership guide point to four:

Atlassian's partnership guide also notes that a first-time connection "may require site-administrator approval." If the consent screen works but nothing comes back, ask your Jira admin to check these settings before you debug anything else.

Admins can also allow or block access by intent: read, write, and search. Each tool inherits the access of its group.

What the Jira MCP server can do

Atlassian's server doesn't dump every tool into your context at once. It exposes a few "primary" tools and a discover tool the agent uses to find the rest. Writes go through executeWrite and deletes through executeDestructive, so clients can ask for confirmation at the right level.

The Jira tools Atlassian lists in its supported-tools reference include:

Confluence has matching read and write groups, which is why "atlassian mcp server" and "confluence mcp" searches lead to the same setup. Connect once and you get both.

Prompts that work well

JQL search

Find all open bugs in project PAY assigned to me, sorted by priority. Show me the JQL you used.

Asking for the JQL lets you check the query. It's also a quick way to learn JQL you can reuse in a saved filter.

Create an issue

Create a Story in project PAY: "Add retry to webhook sender". Put the acceptance criteria from this conversation in the description and link it to PAY-412.

Transition an issue

Move PAY-456 to In Review and add a comment that the PR is up.

That last one is almost word for word an example Atlassian uses in its own docs.

Cross-product

Summarize the Q4 planning page in Confluence, then create one Jira task per action item and assign them to the owners named on the page.

What it costs to run

There's no separate price for the server, and not every call costs credits. Atlassian's Rovo usage table splits MCP usage into two groups. Free: installing the MCP server, requesting single-product context such as looking up one Jira work item or Confluence page, and write operations such as updating a Jira work item. Credit-consuming: enriched calls that run complex queries, cross-product traversals or unified search and context tools (Atlassian's example is getTeamworkGraphContext), drawn from the same organizational pool as Rovo Chat, Studio, and Agents. So ordinary Jira lookups and updates shouldn't touch your allowance, while broad searches across products will. Thresholds and plan allowances are on Atlassian's Rovo usage limits page. Check it rather than trusting any number in a blog post.

For Jira Data Center or Server: mcp-atlassian

If your Jira runs on your own infrastructure, Atlassian's hosted server can't reach it. The most widely used alternative is mcp-atlassian, an open-source Python server.

From its README:

• License: MIT. The README states it's "Not an official Atlassian product."
• Compatibility: Jira Cloud (fully supported) and Jira Server/Data Center v8.14+. Confluence Cloud and Server/Data Center v6.0+.
• Auth: API token for Cloud. Personal Access Token for Server/Data Center.
• Tools: jira_search (JQL), jira_get_issue, jira_create_issue, jira_update_issue, jira_transition_issue, plus Confluence equivalents.

It runs locally with uvx. In Claude Code, using the standard --env syntax from the Claude Code MCP docs, a Data Center setup looks like this (replace the URL and token with your own):

claude mcp add --transport stdio \
  --env JIRA_URL=https://jira.your-company.com \
  --env JIRA_PERSONAL_TOKEN=your_personal_access_token \
  --env READ_ONLY_MODE=true \
  jira -- uvx mcp-atlassian

READ_ONLY_MODE=true disables write operations. Start there, confirm search works, then remove it when you actually want the agent creating and moving issues. The project's docs also support ENABLED_TOOLS and TOOLSETS to expose only the tools you need.

Security: tokens, write access, and prompt injection

Atlassian's own disclaimer is the right starting point: "MCP clients can perform actions on all connected products (such as Jira, Confluence, Bitbucket) with your existing permissions. Use least privilege, review high‑impact changes before confirming, and monitor audit logs for unusual activity."

In practice:

• Prefer OAuth over tokens. Atlassian recommends API tokens only for non-interactive cases like CI. Token auth also skips domain allowlist checks; only your IP allowlist applies.
• Treat a token like a password. A Jira API token or PAT acts as you. Don't commit it in .mcp.json; keep it in an environment variable or secrets manager, and rotate it when people leave.
• Leave delete off. Atlassian disables delete_jira by default for a reason. Issue deletion can't be undone.
• Watch for prompt injection. Jira tickets and Confluence pages are text anyone in your org, and sometimes customers, can write. A ticket description can contain instructions aimed at your agent. Review write calls before approving them. Our MCP security risks post covers the attack in detail.

If you're comparing Jira with other servers to add to your setup, our best MCP servers guide has the checks we use before installing any of them.

FAQ

Is there an official Jira MCP server?

Yes. Atlassian runs the Atlassian MCP Server (also called the Atlassian Rovo MCP Server in some of its docs) at https://mcp.atlassian.com/v2/mcp. It covers Jira, Confluence, and other Atlassian Cloud apps.

Does the Atlassian MCP server work with Jira Data Center?

Atlassian's setup docs require an Atlassian cloud site. For Data Center or Server, use an open-source server such as mcp-atlassian, which supports Jira Server/Data Center v8.14+.

Is the Atlassian MCP server free?

Atlassian doesn't charge separately for the connection. Single-product lookups (a Jira work item, a Confluence page) and write operations are free; enriched, cross-product and unified search/context calls consume Rovo credits from your organization's shared pool. See Atlassian's Rovo usage limits page for allowances.

Is there a separate Confluence MCP server?

Not from Atlassian. The same server covers Confluence; tools like getConfluenceContent and createConfluenceContent sit alongside the Jira tools.

Why can I sign in but get no data back?

Usually an admin setting: your IP isn't on the org's allowlist, partner domains are blocked, or you lack permission on that project. Atlassian's OAuth guide lists "token expired or improperly scoped" as another cause; re-run /mcp to re-authenticate.

Can the agent delete Jira issues?

Not by default. The delete_jira tool group is disabled until an admin enables it.

Wrapping up

• On Atlassian Cloud, use the official server: https://mcp.atlassian.com/v2/mcp, one claude mcp add command, OAuth sign-in.
• It only does what your Jira account can do, and admins can restrict it further.
• Single-product lookups and writes are free; enriched and cross-product calls consume Rovo credits. Check your plan's allowance.
• On Data Center or Server, use mcp-atlassian with a PAT, and start in read-only mode.
• Review every write, and leave delete off.

If the setup step is where you keep stalling, Taku mirrors a working AI setup into a desktop workspace and runs it, so you can start from a configuration someone already got working. Taku is in Beta, and the Mac app is available now.