Jira for coding agents
Jira tickets for your coding agent, without configuring an MCP server.
An administrator connects Jira Cloud once for the workspace. Open a ticket in the Mac app and start a session: your agent gets the ticket and Kanjin’s own MCP server, with nothing to install and no file to change.
The ticket, in the first message
A ticket session starts with a brief: the key, the summary, the type, the status, the description and the comments; the repositories, branches and pull requests that name the ticket; the local paths and worktrees; and the Confluence and Notion pages that the ticket links to.
The brief tells the agent that the ticket text comes from the issue tracker, and that it is data, not instructions.
One server for each session
The Kanjin MCP server runs inside the Mac app, on 127.0.0.1, with a new token for each session. The token lives only in the agent’s environment.
Claude Code gets the server with --mcp-config, Codex in the configuration of its thread, and opencode in OPENCODE_CONFIG_CONTENT. No configuration file changes.
Your account, your access
An administrator connects Jira Cloud with Atlassian’s own authorization. Each member links their own Atlassian account, and each read uses that account, so the agent sees only the tickets that you can see.
Every change waits for you
A status change, a ticket comment and new worktrees each show an approval card with the exact text, also inside a chain or an automation. A tool works only on the ticket of its session.
The tools
- read_ticketThe ticket, its comments and its status
- read_knowledge_pageA Confluence or Notion page, or a vault note, as Markdown
- classify_changesWhat the classifier makes of a worktree’s changes
- create_worktreesThe worktrees, in the repositories that the agent names, after you approve them
- set_ticket_statusA new status, after you approve it
- draft_ticket_commentA comment on the ticket, after you approve it
- draft_slack_messageA message to a thread, a channel or a person that the agent found, after you approve it
- start_local_runThe commands that start each worktree, after you approve them, and their output
Compared with a Jira MCP server in your agent’s settings
You can add a Jira MCP server to each agent yourself. Then each agent needs its own setup and its own token, and the agent can reach every ticket that your account can reach.
Kanjin’s server comes with the session: one ticket, your own account, and an approval for each change. The same session also gets worktrees, the steps of the ticket and a review of its changes.
Jira in the rest of Kanjin
The Work area shows boards, sprints, filters, epics and pinned tickets, from as many Jira sites as the workspace uses. An automation runs a saved chain when a ticket enters a Jira list, or on a schedule.
Questions
Which agents get the Jira tools?
Claude Code, Codex and opencode. Gemini CLI gets the ticket brief as its first prompt, without the tools.
Which Jira works?
Jira Cloud, through Atlassian’s own authorization. A workspace can connect many Jira Cloud sites, and Linear workspaces as well.
Does my agent still get my own MCP servers?
Yes. In a ticket session, Claude Code, Codex and opencode also load your own MCP servers, with your accounts.
Can I use the Kanjin server in another tool?
No. It runs inside the Mac app, for Kanjin’s own sessions, and a tool works only on the ticket of its session.
What does Kanjin keep from Jira?
Kanjin reads tickets and comments live, with your account, and does not store them. It keeps the account ID, the email address and the display name of an Atlassian account for 24 hours at most.
Does Linear work the same way?
Not yet. Linear lists issues, with search, pins and status changes. A Linear issue has no ticket session yet.