Jira MCP服务器

@khanglvm/jira-mcp

MCP server for legacy Jira Server (v7.x) with Basic Authentication. Works with any MCP-compatible AI tool.

Using Jira Cloud or Data Center 8.14+? Use mcp-atlassian instead for OAuth/PAT support.

Prerequisites

Node.js (v18+) is required.

macOS / Linux:

curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.1/install.sh | bash

Windows (PowerShell):

winget install -e --id OpenJS.NodeJS.LTS

Quick Install

Install using mcpm (MCP Manager) — a universal installer that works with all AI agents:

npx @khanglvm/mcpm https://github.com/khanglvm/jira-mcp

What this does:

• Auto-detects your installed AI agents (Claude Desktop, Cursor, VS Code, etc.)
• Prompts for your Jira credentials securely
• Transforms config to match each agent's format
• Safely merges with existing configurations

AI Agents: Do NOT use this method — it requires interactive terminal input. See For AI Agents below.

For AI Agents

MCP Installation

Follow the MCP Server Installation Workflow for automated setup:

• Automatic agent detection and configuration
• Secure credential collection
• Proper config format transformation for your specific agent

Agent Skill (Recommended)

Install the companion skill for JQL expertise, gotcha prevention, and response formatting:

# Install globally (available across all projects)
npx skills add https://github.com/khanglvm/skills --skill jira-mcp -g

# Install for a specific agent
npx skills add https://github.com/khanglvm/skills --skill jira-mcp -a claude-code -g

# Headless install (non-interactive, for CI/automation)
npx skills add https://github.com/khanglvm/skills --skill jira-mcp -a claude-code -g -y

The skill provides:

• Complete JQL reference — operators, fields, functions, and date syntax
• Critical gotchas — statusCategory ("To Do") vs status vs type (Bug/Task)
• Common patterns — ready-to-use JQL queries for standup prep, sprint overview, bug triage
• Workflow guidance — transition-based status changes, self-hosted URL construction
• Response formatting — clickable ticket links, table layouts, status icons

Available Tools

jira_get_release_board accepts links such as https://jira.example.com/projects/PROJ/versions/12345. It returns the version, all directly assigned issues, hydrated Jira subtasks, status counts, and completeness evidence in one call. It preserves tickets outside the caller's scope so product tooling can classify ownership explicitly.

Environment Variables

Best Practices

JQL Query Tips

• Use statusCategory for broad filtering ("To Do", "In Progress", "Done") — NOT status or type
• Always quote values with spaces: project = "My Project"
• Use ORDER BY to sort results: ORDER BY created DESC
• Specify fields in jira_search to reduce response size: ["summary", "status", "assignee"]

Workflow Transitions

• You cannot set status directly — use jira_get_transitions to get valid transition IDs, then jira_transition_issue
• Always call jira_get_current_user first to verify authentication

Common Gotcha

type = "To Do"  → ERROR: "The value 'To Do' does not exist for the field 'type'"

"To Do" is a status category, not an issue type. Use statusCategory = "To Do" instead.

Attachments

• jira_list_attachments returns each attachment's id, filename, mimeType, size, created, author, content (download URL) and thumbnail. Metadata is read from the issue's fields.attachment[] via the legacy GET /rest/api/2/issue/{key}?fields=attachment endpoint.
• jira_get_attachment accepts an attachmentId, or resolves an attachment by issueKey + filename. Images (mimeType starting with image/) are returned as a viewable MCP image content block; other files are written to os.tmpdir()/jira-mcp/<id>-<filename> and returned as a text block with the saved path and metadata (inline base64 is included only when the file is small, under 256KB).
• The attachment content URL is an absolute URL outside the /rest/api/2 base; bytes are fetched with a raw authenticated request that reuses the Basic auth header and follows redirects. This is a legacy Jira Server v7.x integration — only /rest/api/2 endpoints are used (no Cloud /rest/api/3 or ADF).

Changelog

v1.7.1

• security: omit issue descriptions from aggregate release-board results so embedded credentials or private payloads do not enter bulk agent output.

v1.7.0

• feat: add jira_get_release_board with URL validation, paginated release membership, hydrated subtasks, counts, and completeness evidence.
• fix: report the package version through MCP server metadata and --version.
• test: run release-board and null-field offline contracts before publishing.

v1.6.1

• fix: null-guard all issue/search/transition/comment field rendering. jira_get_issue and jira_search no longer throw Cannot read properties of undefined (reading 'name') when a ticket has a null/absent assignee, priority, status, issuetype, project, transition to/statusCategory, or comment author. Missing values now normalize to null.
• test: add test:null regression test covering null/missing nested fields.

v1.6.0

• feat: add attachment access tools jira_list_attachments and jira_get_attachment (images returned as inline image content blocks; other files saved to a temp path). Legacy Jira Server v2 API only.

v1.5.0

• feat: migrate agent skill to khanglvm/skills for global installation via npx skills
• docs: rewrite README with best practices, updated skill installation guidance
• chore: remove .agent directory (skill now lives in dedicated skills repo)

v1.4.0

• feat: add mcp.json for mcpm tool support
• feat: improve tool descriptions with JQL gotchas and add AI agent skill
• fix: add .mjs extension for Node.js ESM compatibility in temporary files
• docs: add mcpm quick install instructions and AI agent skill reference

License

MIT