Kanjin for Claude Code
Claude Code, from a ticket to a pull request.
Kanjin runs Claude Code on your Mac, with your own sign-in and plan. It brings the ticket, a worktree in each repository, its own tools, and a diff that folds the mechanical edits. Claude Code writes the code, and you review it.
A window for Claude Code
An agent session shows one Claude Code session in two views. The Terminal view is Claude Code’s own interface, with all its keys, commands and approvals. The Stream view shows its messages and its approval requests beside every change since the session started.
At the end of each turn, the classifier folds the blocks that preserve behavior, and a Last turn mark shows what the turn changed. A session starts in any folder; the Stream view of the changes needs a Git repository.
Jira tickets in Claude Code, without an MCP setup
Open a Jira ticket and ask Claude Code about it. The first message holds a brief: the description and the comments, the repositories and pull requests that name the ticket, and the pages that it links to.
Kanjin passes its own MCP server to Claude Code on the command line, with a new token for each session. Claude Code reads the ticket and its Confluence or Notion pages, and drafts a comment or a status change for your approval.
A git worktree for each ticket
Prepare Worktrees chooses the repositories that the ticket needs, then creates the branch feature/<KEY> and a worktree in each. Kanjin copies files such as .env and runs each repository’s setup command.
Until the worktrees exist, Claude Code gets only its read tools for your files. In Implement and Fix, its Bash sandbox and its file tools reach only the ticket’s worktrees, and no approval can widen that. Your main checkout stays as it is.
Claude Code review, before the pull request
Review reads the changes of every worktree and lists defects. Overengineer looks for needless layers and unclear names. Fix checks each finding against the code and the ticket, fixes the valid ones and skips the rest. It does not commit.
Review and Overengineer can also run on a second agent, such as Codex, and report back to the Claude Code session. Then you read the diff, with the mechanical edits folded.
Your plan, your tools
Kanjin builds no agent of its own. It runs the Claude Code that you installed, with your own sign-in, so a Claude plan works as it does in your terminal.
In a ticket session, Claude Code also gets the web and your own MCP servers. These tools work outside the sandbox, with your accounts, and the start panel says so.
Sessions that stay, and chains that run
A ticket session stays when you open another item, and it continues on the same Claude Code conversation after Kanjin restarts.
A chain runs up to eight steps of a ticket in order. An automation starts a saved chain when a ticket enters a Jira list, and every approval still waits for you.
Questions
Is Kanjin a GUI for Claude Code?
In part. Kanjin shows Claude Code’s own terminal and a Stream view of the same session. It is also a development environment around it: Jira tickets, git worktrees, the steps of a ticket, and a review of each change.
Do I need an Anthropic API key?
No. Kanjin starts the claude command that your login shell finds, with the sign-in that Claude Code already has, such as your Claude plan. Kanjin holds no model keys and bills no tokens.
Does Kanjin change my Claude Code settings?
No. Kanjin passes its MCP server and the limits of a session as options of the claude command. It changes no configuration file.
Can Claude Code commit or push from Kanjin?
Implement and Fix never commit. When you start Commit & Push, Claude Code commits in each worktree, and Kanjin pushes only after you confirm the plan for each repository.
Can one ticket use Claude Code and Codex?
Yes. The ticket session runs on one agent, and Investigate, Review and Overengineer can run on a second one. The second agent starts read-only and reports back to the session.
Is Kanjin made by Anthropic?
No. Kanjin is an independent product. Claude Code is a product of Anthropic.