Jira MCP
Enables AI clients to manage Jira Cloud issues, comments, attachments, and transitions through natural language.
README
Jira MCP
A local Model Context Protocol (MCP) server for Jira Cloud. It lets AI clients such as Codex and Claude search issues, inspect descriptions and comments, create or update issues, add rich comments, transition statuses, and read attachments.
Jira access uses an Atlassian account email and API token. Credentials stay on your machine in ~/.jira-mcp/sites.json; the server binds to 127.0.0.1 by default.
Features
- Streamable HTTP MCP endpoint with legacy SSE compatibility
- Local Web UI for configuring multiple Jira Cloud sites
- JQL search using Jira Cloud's enhanced
/rest/api/3/search/jqlAPI - Issue creation, updates, comments, rich ADF comments, and transitions
- Attachment discovery and bounded inline image downloads
- Tool and client activity log
- Write-only API tokens in the Web UI/API
- User-only config directory and file permissions
- Local MCP OAuth compatibility endpoints for clients that require discovery
Requirements
- Node.js 22.14 or newer
- A Jira Cloud site
- An Atlassian account email and API token from Atlassian account security
Installation
Run with npx
npx --yes @yunusemregul/jira-mcp@latest
Install globally
npm install --global @yunusemregul/jira-mcp
jira-mcp
The Web UI opens at http://127.0.0.1:18433. Add a Jira site, email, and API token there.
Options:
-p, --port Port to listen on (default: 18433, env: PORT)
--host Host to bind (default: 127.0.0.1, env: HOST)
-v, --version Print version
-h, --help Show help
Connect an MCP client
Codex
codex mcp add jira-mcp --url http://127.0.0.1:18433/mcp
Or add this to ~/.codex/config.toml:
[mcp_servers.jira-mcp]
url = "http://127.0.0.1:18433/mcp"
Claude Code
claude mcp add --transport sse jira-mcp http://127.0.0.1:18433/mcp/sse
Other Streamable HTTP clients
{
"mcpServers": {
"jira-mcp": {
"url": "http://127.0.0.1:18433/mcp"
}
}
}
Some MCP clients require OAuth discovery even for a local server. Jira MCP exposes the same auto-approve compatibility endpoints as hac-mcp. They do not authenticate or authorize Jira access; the loopback bind is the security boundary.
Available tools
| Tool | Description |
|---|---|
list_sites |
List configured Jira sites without exposing credentials |
get_projects |
List or filter accessible Jira projects |
search_issues |
Search issues with JQL and token-based pagination |
get_issue |
Read issue details, comments, custom text fields, and attachments |
create_issue |
Create an issue |
update_issue |
Update summary, description, priority, assignee, or labels |
add_comment |
Add a plain-text comment |
add_comment_rich |
Add an Atlassian Document Format comment |
get_comments |
Read comments |
update_comment |
Replace an existing comment |
get_attachments |
List issue attachments and referenced media IDs |
read_attachment |
Read a bounded attachment; images are returned inline |
get_transitions |
List available status transitions |
transition_issue |
Transition an issue, optionally with a comment |
Configuration
The CLI accepts these optional environment variables:
| Variable | Default | Purpose |
|---|---|---|
HOST |
127.0.0.1 |
HTTP bind host |
PORT |
18433 |
HTTP port |
JIRA_REQUEST_TIMEOUT_MS |
30000 |
Jira HTTP request timeout |
JIRA_MAX_ATTACHMENT_BYTES |
20971520 |
Maximum attachment download size |
JIRA_MCP_DATA_DIR |
~/.jira-mcp |
Config directory; primarily useful for tests/isolated runs |
Security
- Keep the default loopback bind. Setting
HOST=0.0.0.0exposes an unauthenticated local MCP service and Web UI; only do this behind a trusted authenticated reverse proxy. - API tokens are stored locally in plaintext because the server must use them. The config directory is forced to mode
0700andsites.jsonto0600on POSIX systems. - API responses never return saved tokens. When editing a site, leave the token field blank to retain the existing token.
- Jira site URLs must use HTTPS and cannot embed credentials.
- The local OAuth compatibility endpoints auto-approve clients and are not a security mechanism.
- Attachment downloads default to a 20 MiB limit.
See SECURITY.md for reporting vulnerabilities.
Development
npm ci
npm run verify
npm start
npm run verify runs ESLint and the Node test suite. npm publish runs the same checks automatically through prepublishOnly.
Publishing
The npm package name is @yunusemregul/jira-mcp. The first publish must be performed interactively after enabling npm 2FA:
npm login
npm run verify
npm publish --access public
After the package exists, configure npm Trusted Publishing for .github/workflows/publish.yml. Subsequent v<package-version> tags publish from GitHub Actions using OIDC and npm provenance without a long-lived write token.
License
MIT License with the Commons Clause condition. See LICENSE.
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。