jira-dev-mcp
A local MCP server for Jira Cloud-driven development. Connects MCP clients to Jira for searching issues, reading tasks, parsing attachments, and mapping projects to local repos.
README
jira-dev-mcp
A local MCP server for Jira Cloud-driven development. Connects Claude Code, OpenCode, Codex, and other MCP clients to your Jira instance — search issues, read task details, parse attachments, and map Jira projects to local repositories.
Features
- Search Jira issues with natural language or raw JQL
- Read full task details: description, subtasks, changelog, comments, attachments
- Download and parse CSV, XLSX, XLS, PDF attachments via Python
- Images and binary files passed to AI client as base64
- Map Jira project keys to local repository paths
- OAuth 2.0 (3LO) browser-based login — no manual token management
- Auto-refresh OAuth tokens before expiry
- Post comments on Jira issues with clickable URL response
- Token-only comment confirmation: confirm pending comments with just the token, no need to re-send body
- Edit existing comments with the same confirmation flow
- Batch download all attachments from an issue in a single call with optional MIME filter
- Auto-retry on transient API errors (429/503) with Retry-After header support
- Rich ADF parsing: code blocks, tables, mentions, links, panels, and more
- Multi-paragraph and fenced code block support in posted comments
- Typed error hierarchy for programmatic error handling
- OAuth callback port auto-fallback when default port is busy
- Read and write Jira access
Install
npm install -g jira-dev-mcp
Authentication
Option A: OAuth 2.0 (Recommended)
Just run:
jira-dev login
A browser window opens for Jira authorization. Tokens are saved to ~/.jira-dev/config.json and auto-refreshed. No manual Client ID or Secret required.
Advanced: To use your own OAuth app, set
JIRA_CLIENT_IDandJIRA_CLIENT_SECRETbefore runningjira-dev login.
Option B: API Token (Basic Auth)
export JIRA_BASE_URL="https://your-domain.atlassian.net"
export JIRA_EMAIL="you@example.com"
export JIRA_TOKEN="your-jira-api-token"
Optional — store token in macOS Keychain instead of env var:
security add-generic-password -a "$USER" -s "jira-dev-mcp:JIRA_TOKEN" -w "your-token"
MCP Client Configuration
Auto-register (easiest)
jira-dev setup
This creates or updates ~/.claude.json (Claude Code) and ~/.opencode/config.json (OpenCode) automatically.
If an existing config file contains invalid JSON, jira-dev setup now fails with an explicit error instead of silently skipping it.
Manual — Claude Code (~/.claude.json)
{
"mcpServers": {
"jira": {
"command": "jira-dev",
"args": ["server"]
}
}
}
Manual — OpenCode / Codex
{
"mcpServers": {
"jira": {
"command": "jira-dev",
"args": ["server"]
}
}
}
With environment variables
{
"mcpServers": {
"jira": {
"command": "jira-mcp-server",
"env": {
"JIRA_BASE_URL": "https://your-domain.atlassian.net",
"JIRA_EMAIL": "you@example.com",
"JIRA_TOKEN": "your-token"
}
}
}
}
Tools
| Tool | Description |
|---|---|
jira_search_issues |
Search by keywords or JQL |
jira_read_task |
Read full issue details |
jira_download_attachment |
Download and parse a single attachment |
jira_download_all_attachments |
Download all attachments in one call; optional MIME filter (e.g. image/) |
jira_my_tasks |
List issues assigned to you |
jira_add_comment |
Post a comment; confirm with token only (no body re-send needed) |
jira_edit_comment |
Edit an existing comment; same token-only confirm flow |
jira_set_project_path |
Map a Jira project to a local repo path |
jira_get_project_path |
Get the local path for a project |
jira_analyze_task |
Full investigation workflow: reads issue + comments + attachments, selects type-aware template (Bug/Story/Task), and guides step-by-step through explore → plan → implement → post comment |
CLI Commands
jira-dev status # Show current auth mode, token state, and mapped projects
jira-dev doctor # Run environment and config health checks
jira-dev upgrade # Upgrade jira-dev-mcp from npm
jira-dev setup # Register jira-dev into supported MCP clients
jira-dev read AT-123 # Read issue details from the terminal
jira-dev comment AT-123 "Done" # Post a comment directly (bypasses manual mode)
jira-dev download AT-123 # Download all attachments
jira-dev download AT-123 spec.xlsx # Download a single attachment
jira-dev config set-comment-mode manual # Require confirmation before posting comments
Development Workflow
Standard usage with Claude Code
Step 1 — One-time setup: map your project
jira_set_project_path(jiraProject: "AT", localPath: "/path/to/your/repo")
Step 2 — Find your task
jira_my_tasks(status: "In Progress")
or search by keyword / JQL:
jira_search_issues(query: "login timeout bug")
jira_search_issues(query: "project = AT AND sprint in openSprints()")
Step 3 — Read the task
jira_read_task(key: "AT-123", includeComments: true)
Returns: description, subtasks, changelog, comments, attachment list, and the local repo path.
Step 4 — Download attachments if needed
jira_download_all_attachments(key: "AT-123")
jira_download_all_attachments(key: "AT-123", mime_filter: "image/")
jira_download_attachment(key: "AT-123", filename: "spec.xlsx")
Use jira_download_all_attachments to grab everything in one call, or filter by MIME type. Use jira_download_attachment for a single file. CSV / XLSX / XLS / PDF are parsed and returned as structured text. Images are returned as base64.
Step 5 — AI implements the fix
With the task details and local repo path in context, ask Claude Code to:
- Explain the root cause
- Propose a plan with impact scope
- Implement the fix
- Write test cases
Project Path Mapping
Map a Jira project key to a local repo so the AI knows where to look:
jira_set_project_path(jiraProject: "AT", localPath: "/Users/you/projects/my-app")
Python Dependency (Attachment Parsing)
Python 3 is required for CSV, XLSX, XLS, and PDF parsing. For XLS and PDF, additional packages are auto-installed on first use, or install manually:
python3 -m pip install -r $(npm root -g)/jira-dev-mcp/scripts/requirements.txt
Security
- OAuth tokens stored at
~/.jira-dev/config.jsonwith600permissions - API tokens: prefer env vars or macOS Keychain over config file
- Attachment size and MIME type allowlist enforced before download
- OAuth client credentials injected at build time — never committed to source
- Secrets are never written to logs
Local Development
git clone https://github.com/nianyi778/jira-dev-mcp.git
cd jira-dev-mcp
npm install
npm run generate-defaults # required before first build (uses env vars or empty placeholders)
npm test
npm run build
Inspect tools interactively:
npx @modelcontextprotocol/inspector node dist/index.js
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。