jira-mcp
MCP server for Jira Cloud enabling issue tracking, comments, transitions, and attachments management through natural language.
README
jira-mcp
MCP server for Jira Cloud — issue tracking, comments, transitions, attachments.
Install
{
"mcpServers": {
"jira": {
"command": "uvx",
"args": ["--refresh", "--extra-index-url", "https://nikitatsym.github.io/jira-mcp/simple", "jira-mcp"],
"env": {
"JIRA_URL": "https://myteam.atlassian.net",
"JIRA_EMAIL": "you@example.com",
"JIRA_TOKEN": "your-api-token",
"MCP_JIRA_BRIEF_MAX": "100"
}
}
}
}
Get an API token at https://id.atlassian.com/manage-profile/security/api-tokens
Groups
| Tool | Description |
|---|---|
jira_read |
Search, list projects/boards/sprints, fetch issues/comments/transitions (read-only) |
jira_write |
Create/update issues, comments, links, worklogs, attachments (non-destructive) |
jira_execute |
Trigger state changes: transition, assign, unassign |
jira_delete |
Delete issues, comments, links, attachments (destructive) |
Call any group with operation="help" to list operations and their typed signatures; operation="schema" returns JSON Schema for one op (params={"op": "OpName"}).
Environment
| Var | Required | Default | Purpose |
|---|---|---|---|
JIRA_URL |
yes | — | Atlassian site URL, e.g. https://myteam.atlassian.net |
JIRA_EMAIL |
yes | — | Atlassian account email (basic-auth pair with token) |
JIRA_TOKEN |
yes | — | API token from id.atlassian.com |
MCP_JIRA_BRIEF_MAX |
no | 100 |
Max length of <brief> summary on issue descriptions. 0 disables enforcement |
Gotchas
parent_key vs Epic Link customfield. Team-managed (next-gen) projects accept parent_key="STS-1" and turn it into fields.parent={"key": "STS-1"}. Company-managed (classic) projects with Epic Link enabled require custom_fields={"customfield_NNNNN": "STS-1"} where NNNNN is the Epic Link field id (varies per site — see jira_read.ListFields). parent_key will not error in this case but the link won't be drawn either; the brief idea is to use custom_fields when working with classic projects.
ADF. description, environment, and comment bodies are converted from plain text to Atlassian Document Format. No markdown is rendered — *bold* stays literal. Roundtrip is lossy (the body returned by Jira is normalized ADF, not your input text), so _verify_top_level skips body from the verify pass.
accountId vs username/email. Jira Cloud accepts only accountId (e.g. 5e9c3a99...) for assignee/reporter/etc. Email or username will be silently rewritten or rejected. Look up accountIds with jira_read.SearchUsers.
<brief> requirement on create_issue.description. When MCP_JIRA_BRIEF_MAX > 0 (default 100), the description must contain <brief>one-line summary</brief>. Set MCP_JIRA_BRIEF_MAX=0 in env to disable. Brief is enforced on description only (not on comment bodies).
Verify-writes. After every write that returns a resource, jira-mcp does either zero extra RTT (update_issue uses ?returnIssue=true, comment ops use the POST response) or one slim follow-up GET (create_issue/transition_issue/assign_issue/unassign_issue, with ?fields=<sent keys>). The verify pass catches dropped fields — Jira silently ignored what we sent. It does NOT catch substituted values — if Jira swaps an invalid priority="Highest" for the project default, the key is present but the value differs. Strict-equality checks (assignee accountId, unassign null) are done inline as separate raises.
Resolution on update. update_issue rejects resolution in custom_fields because most workflow screens drop it silently. Use jira_execute.TransitionIssue(..., resolution=...). If the workflow screen still rejects the fields.resolution shape with a 400, retry passing the value via the update= parameter: update={"resolution": [{"set": {"name": "Done"}}]}.
upload_attachment reads any local file. The op passes file_path straight to open(...). An agent with this tool can read and POST any file the MCP process can read. Treat reach as equivalent to the process's filesystem scope; do not run jira-mcp with elevated privileges or wider filesystem access than you'd grant the model.
download_attachment path-traversal hardening. The path is None branch sanitizes the server-supplied filename via os.path.basename(...) or f"attachment_<id>", caps it at 255 chars, and saves into a fresh per-call mkdtemp("jira_mcp_") directory. Explicit path=... is the caller's responsibility.
Development
Tested against a real Jira Cloud sandbox. Unit tests use httpx.MockTransport (no network); integration tests are gated on env vars.
uv sync
npm run test # unit tests, no network
npm run jira:bootstrap # verify tests/.env against your sandbox
npm run test:integration # end-to-end against real Jira
tests/.env (gitignored) must contain JIRA_URL/JIRA_EMAIL/JIRA_TOKEN/JIRA_TEST_PROJECT. CI runs the integration job against the same secrets when set on the repo.
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。