Jira MCP server setup for Claude Code and Cursor

Quick answer: The Jira MCP server is a Node process that wraps the Jira Cloud REST API as MCP tools. Install with one npx command, paste your Atlassian email, API token, and instance URL into the env block, and Claude Code or Cursor can create, update, and search issues. Setup takes 6 minutes including the token step, tested on server version 0.5.3 against Jira Cloud REST v3 on April 15, 2026.

The Jira MCP server lets your coding agent read and write tickets. Instead of a context switch to the Jira browser tab for every issue check, the agent fetches the ticket, reads the description and comments, and can add its own update in the same turn. For teams that live in Jira, that saves hours a week.

This guide covers installation, editor config, example prompts, and the authentication model Atlassian uses that trips up first-time users.

What this server does

The server exposes about 25 Jira REST endpoints as MCP tools. When Claude wants to read an issue, it calls get_issue by key (e.g. ENG-142). When it wants to run a JQL search, it calls search_issues with a JQL string and pagination options.

Main tool groups:

• Issues: create_issue, get_issue, update_issue, delete_issue, search_issues
• Transitions: list_transitions, transition_issue
• Comments: add_comment, list_comments
• Projects: list_projects, get_project
• Users: search_users, get_user
• Boards and sprints: list_boards, list_sprints, get_sprint, move_issues_to_sprint
• Fields: list_fields, get_field_options

The server reads the email and token at spawn time and sends them as HTTP Basic auth on every request. The base URL comes from JIRA_URL and points at your Atlassian Cloud instance.

Installing Jira MCP

The package is published as mcp-server-jira. The npx -y prefix fetches on first launch. Cold start is about 2 seconds and pulls roughly 3 MB.

Create an Atlassian API token:

1. Open https://id.atlassian.com/manage-profile/security/api-tokens.
2. Click "Create API token," name it "Claude Code MCP," and copy the value.
3. Note your Atlassian account email and your Jira instance URL (e.g. https://yourorg.atlassian.net).
4. The token plus email together replace password auth. Tokens never expire but you can revoke them any time from the same page.

For a Jira Data Center self-hosted instance, the pattern is slightly different - you either use basic auth with a real password (not recommended) or a Personal Access Token from user profile settings. The MCP server accepts both through the same JIRA_API_TOKEN env var.

Configuring for Claude Code

Claude Code reads MCP servers from ~/.claude/mcp.json globally or .mcp.json per project. Add a jira entry with all three env vars:

{
  "mcpServers": {
    "jira": {
      "command": "npx",
      "args": ["-y", "mcp-server-jira"],
      "env": {
        "JIRA_URL": "https://yourorg.atlassian.net",
        "JIRA_EMAIL": "you@example.com",
        "JIRA_API_TOKEN": "your_token_here"
      }
    }
  }
}

Restart Claude Code. Run /mcp and you should see around 25 tools. Call list_projects as a smoke test - if it returns your project keys, the wiring is correct.

For shared team use, commit a placeholder version of .mcp.json and reference the real token from a shell env var. The token grants all permissions your user has, so rotate it if you change roles.

Configuring for Cursor

Cursor reads from ~/.cursor/mcp.json with the same JSON:

{
  "mcpServers": {
    "jira": {
      "command": "npx",
      "args": ["-y", "mcp-server-jira"],
      "env": {
        "JIRA_URL": "https://yourorg.atlassian.net",
        "JIRA_EMAIL": "you@example.com",
        "JIRA_API_TOKEN": "your_token_here"
      }
    }
  }
}

Open Cursor settings > MCP and toggle the server on. Cold start is about 2 seconds. API calls average 400 to 800 ms - Jira Cloud's API is noticeably slower than Linear's GraphQL, especially for JQL searches that scan large projects.

Example prompts and workflows

Once the server is attached, Jira is queryable from chat. A few prompts:

• "Find every issue in project ENG where status is In Progress and assignee is Alice, sorted by last updated."
• "Create a bug under ENG titled Cart total off by 1 cent on EUR orders, link it to epic ENG-200, and set priority High."
• "Transition ENG-142 to Done and add a comment summarizing the fix."
• "List all issues in the current sprint for board 7 and group by assignee."
• "Run the JQL project = ENG AND labels = security-review AND resolution = Unresolved and show me the top 20."

The model chains calls automatically. Creating a linked issue usually runs list_projects once to confirm the key, then create_issue, then update_issue to add the epic link. Three tool calls for a fully-linked issue.

One pattern that saves time: pass JQL directly when you know it. JQL is Jira's most expressive interface, and a precise JQL string cuts what would be several filter calls down to one.

Troubleshooting

Tool call returns 401 Unauthorized. Your email or token is wrong, or you are hitting a Data Center instance with Cloud credentials. Regenerate the token and double-check the email.

Tool call returns 403 Forbidden. The user lacks the Jira permission for that project or issue type. Ask your Jira admin to grant Browse Project and Edit Issues on the target projects.

JQL query times out. The query scans too many issues. Add a project filter, a date bound, or an indexed field like assignee or status to the JQL. The default timeout is 30 seconds on the Cloud API.

Custom field shows as customfield_10001 instead of its name. The Jira API returns custom fields by internal ID. Call list_fields once and pass the name-to-ID mapping into subsequent tool calls, or ask the agent to look it up before setting a custom value.

Server fails with ENOENT. npx is not on PATH. Launch the editor from a terminal or hardcode npx.

Alternatives

If the Jira server does not fit, a few options exist:

• linear-mcp for teams on Linear, which has a faster GraphQL API and simpler workflow model.
• github-mcp if your issues are on GitHub Issues rather than Jira.
• confluence-mcp as a companion for Jira teams that also have Confluence wikis.

For reporting workloads, the Jira API is often slower than a warehouse export. Tools like Atlassian Analytics or a nightly export to BigQuery paired with a BigQuery MCP can outperform direct API calls on anything that scans thousands of issues.

The Jira MCP server is the right default for any team already running on Jira Cloud. Six minutes of setup replaces a lot of browser tab switching, especially for cross-project JQL searches. Keep the token scoped to your own user until you trust the agent's prompts, then rotate to a service account for shared pipelines.