Jira Easy MCP Server

jira-easy-mcp

Another Model Context Protocol (MCP) server for Jira, without Auth token. Provides tools for complete Jira automation including issues, projects, boards, sprints, comments, worklogs, and more.

Tested on: Jira Server / Data Center 7.13, 8.20, 9.12 and 10.3 (every tool, automated, see Development & Testing)

---

🔑 Why This MCP?

Works with Basic Authentication (username + password) — no API tokens required!

Most Jira MCP servers require OAuth or API tokens, which many organizations restrict or don't allow. This MCP uses HTTP Basic Authentication, so you can connect using just your Jira username and password. Perfect for:

• 🏢 Enterprise environments where admins don't allow API token creation
• 🔒 Self-hosted Jira Server instances without OAuth configured
• ⚡ Quick setup without going through IT approval processes

---

Quick Start

VS Code (Workspace Configuration)

Create .vscode/mcp.json in your workspace root:

{
  "servers": {
    "jira": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "jira-easy-mcp"],
      "env": {
        "JIRA_BASE_URL": "https://your-jira-instance.com",
        "JIRA_USERNAME": "your-username",
        "JIRA_PASSWORD": "your-password",
        "JIRA_PROJECTS_FILTER": "ABC,DEF,XYZ",
        "JIRA_RESPONSE_FORMAT": "JSON"
      }
    }
  }
}

Add to your VS Code settings for accessing in all workspaces.

Note: JIRA_PROJECTS_FILTER and JIRA_RESPONSE_FORMAT are optional. Projects filter limits access to specified projects. Response format can be JSON (default) or TOON (text-oriented notation).

Claude Code

claude mcp add jira \
  -e JIRA_BASE_URL=https://your-jira-instance.com \
  -e JIRA_USERNAME=your-username \
  -e JIRA_PASSWORD=your-password \
  -- npx -y jira-easy-mcp

Codex CLI

Add to ~/.codex/config.toml:

[mcp_servers.jira]
command = "npx"
args = ["-y", "jira-easy-mcp"]
env = { JIRA_BASE_URL = "https://your-jira-instance.com", JIRA_USERNAME = "your-username", JIRA_PASSWORD = "your-password" }

Other clients

Cursor: ~/.cursor/mcp.json or .cursor/mcp.json (use "mcpServers": { ... } wrapper)
Windsurf: Add to your Windsurf MCP settings (use "mcpServers": { ... } wrapper)
Claude Desktop: Edit claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\; use "mcpServers": { ... } wrapper)
Gemini CLI: ~/.gemini/settings.json (use "mcpServers": { ... } wrapper)

All use the same command, args, and env configuration as VS Code above.

Windows: if the client cannot start npx, use "command": "cmd" with "args": ["/c", "npx", "-y", "jira-easy-mcp"].

Tool limits: some clients (notably Cursor) cap how many tools they send to the model, historically around 40 across all servers. This server has 41; turn off the ones you don't need in the client's MCP settings.

Credentials from the OS keychain (optional)

To keep your password out of config files, omit JIRA_USERNAME/JIRA_PASSWORD and store a credential under the service name JiraMCP:

• macOS: security add-generic-password -s JiraMCP -a your-username -w (prompts for the password; on first start macOS asks to allow node to read it)
• Windows: cmdkey /generic:JiraMCP /user:your-username /pass (use /generic, not /add)

If several accounts are stored, set JIRA_USERNAME to choose one. When both JIRA_USERNAME and JIRA_PASSWORD are set, they are used and the keychain is not read. Keychain support uses the optional keytar package; if it can't be installed on your machine, use the environment variables.

---

Common Use Cases

---

Testing with MCP Inspector

The MCP Inspector provides a web UI to test all tools interactively:

npx @modelcontextprotocol/inspector npx jira-easy-mcp

This opens a browser at http://localhost:6274 where you can:

• View all available tools
• Test each tool with custom parameters
• See the raw JSON responses

Note: Put these values inside inspector UI before connecting: JIRA_BASE_URL, JIRA_USERNAME, JIRA_PASSWORD.

---

Custom fields

Jira REST API v2 identifies custom fields by id (customfield_10006). Values are the raw REST payload for that field type. Display names work only when they are unique (exact, case-sensitive) on the relevant screen.

Read (jira_get_issue)

Default response: system fields plus custom fields that currently have a value, each as { "id", "name", "value" }. Comments, worklogs, and attachments are omitted unless you pass include: ["comments"] (and/or worklogs, attachments). Null custom fields are omitted unless customFields: "all" or you pass fields.

fields narrows the custom-field GET (ids like customfield_11205, or unique edit-screen names like Severity). Requested fields are returned even when null. Names come from this issue's expand=names, not from GET /field.

{
  "key": "ABC-123",
  "customFields": [
    { "id": "customfield_10006", "name": "Story Points", "value": 5 },
    { "id": "customfield_10107", "name": "Flagged", "value": { "value": "Yes" } }
  ]
}

Write (jira_update_issue / jira_create_issue)

Pass customFields keyed by id (customfield_10006) or by a unique name from jira_get_edit_meta / jira_get_create_meta. Ids need no extra GET. Names resolve against that issue type's meta; duplicate names fail closed (use the id). Do not call jira_get_fields before every write.

A field listed by jira_get_fields but missing from edit/create meta is not on that screen and cannot be updated.

Search (jira_search)

Default search returns system fields only (no custom fields, no comment/worklog/attachment dumps). fields is additive extras: system keys and customfield_ ids pass through; display names are resolved with a live uncached GET /field. Prefer ids. Ambiguous names fail closed.

Catalog (jira_get_fields)

Use this for JQL clause names, not as a pre-step for every create/update. For fields on an issue type, use jira_get_edit_meta or jira_get_create_meta.

---

Optional Environment Variables

jira_get_issue_images downloads images through Jira's authenticated attachment URLs (/secure/attachment/..., Jira Server/Data Center). It returns up to five images by default (maximum ten), with a 10 MiB limit per image and 20 MiB per response. jira_upload_attachment reads a local file by absolute path on the machine running the server.

---

Troubleshooting

---

Development & Testing

pnpm install
pnpm test              # unit tests + build (any OS, Node 18+)
node scripts/smoke.mjs # MCP handshake check (needs JIRA_BASE_URL set to anything)

Against a real Jira (Docker, Node 20.10+): the helper starts a throwaway Jira, completes setup with Atlassian's public 3-hour timebomb license, and seeds an admin (admin / Jira-IT-admin-1), a second user, and two Scrum projects (IT, OT).

pnpm jira:up 9.12 8080            # or 7.13, 8.20, 10.3 (Jira 10 also starts PostgreSQL)
JIRA_IT_BASE_URL=http://localhost:8080 pnpm test:integration
pnpm jira:down 9.12

pnpm jira:up 9.12 8080 /jira serves Jira under a context path. Each instance needs about 3 GB of Docker memory; run one at a time on small machines. The keychain tests only run where keytar works and no real JiraMCP credential exists, and they remove what they create. CI runs the unit tests on Linux, macOS and Windows (Node 18-24) and the integration suite against every version above.

---

Compatibility

Requires Node.js 18+. CI runs the unit and MCP smoke tests on macOS, Linux and Windows (Node.js 18–24).

You can improve this by reporting issues