Best Jira MCP Servers in 2026: 5 Ways to Run Jira From Claude

Table of Contents

1. What Is a Jira MCP Server?
2. How We Ranked Them
3. The Best Jira MCP Servers for 2026
4. Quick Comparison Table
5. What You Can Actually Do With a Jira MCP
6. Setup: Connecting Jira to Claude
7. Security: API Tokens and Scopes
8. Frequently Asked Questions

What Is a Jira MCP Server?

A Jira MCP server is a small bridge that exposes Jira's REST API to an AI client through the Model Context Protocol. Once connected, your assistant gains a set of Jira tools it can call mid-conversation: run a JQL search, open a ticket, move an issue from In Progress to Done, drop a comment, or pull the current sprint backlog.

The value is the removed friction. Project managers and engineers spend a surprising amount of time on Jira mechanics, filling fields, hunting for the right issue, copying ticket numbers into standups. A Jira MCP collapses that into natural language. You say what you want, the model translates it into the correct API calls, and you stay in the tool you were already working in.

By 2026 the Atlassian MCP ecosystem has matured into two clear tiers: the official Remote MCP Server that Atlassian maintains for Cloud, and a healthy set of community servers that fill gaps the official one does not cover, particularly self-hosted Data Center support and lighter standalone setups.

How We Ranked Them

We scored every Atlassian-related MCP in the Skiln directory across four signals: maintenance (recent commits, responsive maintainers), coverage (how many Jira operations the tool actually exposes), auth model (OAuth and scoped tokens score higher than raw credentials), and deployment fit (Cloud, Data Center, or both). The five below are the ones we would connect to a real Jira instance today.

The Best Jira MCP Servers for 2026

1. Atlassian Official Remote MCP Server

Atlassian's first-party server, built on its Rovo platform. It covers both Jira and Confluence Cloud, authenticates with OAuth, and runs as a hosted remote endpoint, so there is nothing to install locally. Because it honors your existing Atlassian permissions and logs access in your admin console, it is the right default for any company that cares about auditability.

Best for: Teams on Jira Cloud who want the safest, lowest-maintenance option with official support.

Auth: OAuth via the Atlassian consent flow. No API token to manage by hand.

2. mcp-atlassian (Community)

The most widely deployed community Atlassian MCP. It supports Jira and Confluence across both Cloud and Data Center / Server, which is its killer feature for on-prem shops the official server does not reach. Coverage is deep: JQL search, issue CRUD, transitions, comments, worklogs, sprint and board reads, and Confluence page operations.

Best for: Self-hosted Jira teams and anyone who needs one server for both Jira and Confluence.

Install: uvx mcp-atlassian with JIRA_URL, JIRA_USERNAME, and JIRA_API_TOKEN environment variables.

3. Jira MCP (Standalone)

A focused, Jira-only server for teams that do not use Confluence and want a smaller surface area. It handles the core 80 percent: search, create, update, transition, comment, and assign. Fewer moving parts means a faster install and an easier security review.

Best for: Engineering teams that live in Jira but not Confluence, and want a minimal footprint.

Install: Node-based, configured with your site URL, email, and API token. See the listing for the current package name.

4. Jira + Confluence Combined MCP

Several community servers bundle Jira and Confluence behind one MCP so the assistant can cross-reference a ticket and its spec in the same turn, for example, reading a Confluence requirements page before creating the implementation tickets. If your team's planning lives in Confluence and execution lives in Jira, a combined server keeps both in context.

Best for: Product and engineering teams that link specs in Confluence to delivery in Jira.

Install: Varies by project. Most reuse the same Atlassian API token for both products.

5. Jira Data Center / Server MCP

For organizations still on on-prem Jira, a handful of servers target Data Center specifically, using personal access tokens and a custom base URL. These trade some polish for the ability to run entirely inside your network. Pair one with a self-hosted model or a provider with the right data agreement if your tickets contain regulated information.

Best for: Enterprises on Jira Data Center with on-prem or compliance constraints.

Install: Configure the base URL of your Data Center instance plus a PAT with the needed project scopes.

Quick Comparison Table

What You Can Actually Do With a Jira MCP

Once a Jira MCP is connected, a few patterns quickly become daily habits:

• Generate a standup summary. Ask Claude to pull every issue you moved or commented on yesterday and produce a three-bullet update. No more reconstructing your day from memory.
• Triage the backlog in plain language. "Show me unassigned bugs in the PAYMENTS project, priority high or above, created this month." Claude writes the JQL, runs it, and ranks the results.
• Bulk-create tickets from a spec. Paste a requirements doc and have the assistant draft a parent epic plus child stories, then create them all through the MCP with consistent labels and components.
• Transition issues by description. "Move everything I finished this sprint to Done and add a closing comment." The MCP handles the workflow transitions.
• Cross-reference Confluence and Jira. With a combined server, read a spec page, then open the implementation tickets that trace back to it.
• Draft release notes from a fix version. Pull all issues tagged to a version, group them by type, and produce a clean changelog.

These pair naturally with a code workflow. Many teams run a Jira MCP next to the communication MCP servers they already use for Slack and Linear so the assistant can move work across all three systems in one conversation. See also our Linear MCP review if your team is weighing Jira against Linear.

Setup: Connecting Jira to Claude

For the official remote server, you add it as a remote MCP and complete the OAuth consent flow, nothing to install. For a community server such as mcp-atlassian, the local config looks like this in Claude Desktop:

{   "mcpServers": {     "atlassian": {       "command": "uvx",       "args": ["mcp-atlassian"],       "env": {         "JIRA_URL": "https://your-company.atlassian.net",         "JIRA_USERNAME": "you@company.com",         "JIRA_API_TOKEN": "your_api_token"       }     }   } }

For Claude Code, add it from the CLI instead of hand-editing JSON:

claude mcp add atlassian --env JIRA_URL=https://your-company.atlassian.net --env JIRA_USERNAME=you@company.com --env JIRA_API_TOKEN=xxx -- uvx mcp-atlassian

Restart the client, and the Jira tools register automatically. The Skiln Config Generator can bundle a Jira MCP with the rest of your stack into one ready-to-paste config.

Security: API Tokens and Scopes

A Jira MCP acts with your identity, so scope it carefully:

1. Prefer OAuth. The official Atlassian server surfaces consent screens and logs every action in your admin audit trail. That visibility is worth a lot when an agent is touching real tickets.
2. Use a dedicated service account for automation. Do not wire your personal admin token into a server that runs unattended. A scoped service account limits the blast radius if a token leaks.
3. Start in a sandbox project. Point the MCP at a throwaway project first and watch what the agent does before letting it near production boards.
4. Rotate API tokens on a schedule. Atlassian tokens do not expire by default, so put rotation on a calendar and revoke any token that ends up in a config you no longer control.

Frequently Asked Questions

What is a Jira MCP server?

A Jira MCP server is a Model Context Protocol server that connects an AI client like Claude Desktop, Claude Code, or Cursor to your Jira instance. It exposes Jira operations as tools the AI can call: search issues with JQL, create and update tickets, transition statuses, add comments, manage sprints, and read board state. Instead of switching to Jira and clicking through forms, you describe what you want and the assistant does it through the API.

Is there an official Jira MCP server?

Yes. Atlassian ships an official Remote MCP Server (powered by its Rovo platform) that covers both Jira and Confluence Cloud. It uses OAuth, respects your existing Atlassian permissions, and requires no local install because it runs as a hosted remote endpoint. For self-hosted Jira Data Center, you fall back to community servers like mcp-atlassian.

Does the Jira MCP work with Jira Data Center or only Cloud?

It depends on the server. The official Atlassian Remote MCP targets Cloud. The most popular community option, mcp-atlassian, supports both Jira Cloud and Jira Server / Data Center via a base URL plus a personal access token. If you are on-prem, choose a server that explicitly lists Data Center support in its README.

Can a Jira MCP create and close tickets, or only read them?

Most Jira MCP servers support full read and write: create issues, edit fields, transition workflow states, assign users, log work, and add comments. Some servers expose a read-only mode for safer reporting workflows. Always confirm the tool list in the README and start with a non-production project until you trust the agent's behavior.

What permissions does a Jira MCP need?

A Jira MCP acts as you. With an API token it inherits your account's project permissions, so it can only touch projects and fields you can already access. For least privilege, generate a scoped API token, use a dedicated service account for automation, and prefer the official OAuth server which surfaces granular consent screens.

Can Claude write JQL queries for me?

Yes, and this is one of the best reasons to install a Jira MCP. You describe the filter in plain language ("open bugs assigned to me, updated this week, priority high or above") and Claude generates the JQL, runs it through the MCP, and summarizes the results. It is far faster than remembering JQL syntax for ad hoc reports.

Will a Jira MCP leak sensitive project data to the model?

The MCP only sends data the AI explicitly requests, but anything it reads does enter the model context for that turn. Treat it like any other integration: avoid pointing it at projects with regulated data unless your AI provider has the right data agreement, and use the official OAuth server so access is auditable in your Atlassian admin logs.

Where can I browse other Atlassian and project-management MCPs?

Skiln indexes Jira, Confluence, Linear, Trello, and Asana MCP servers alongside 75,000+ other MCP servers, skills, and agents. Browse the full directory at /browse, or jump straight to the Jira listings at /mcp/pulsemcp-jira and /mcp/atlassian.

Last updated: June 19, 2026 · Skiln tracks MCP server releases daily across 13 source registries.