Leftmark / Claude Code kanban
A kanban board for Claude Code that your agents keep
Leftmark is a Mac app, now in beta, that gives Claude Code, Codex and Gemini one shared kanban board. Every request you give an agent becomes a ticket, and the agent moves it from Doing to Done itself. You see every session's work at a glance, without opening a single terminal tab.
Why a board, when Claude Code already has tasks?
Claude Code's tasks track the steps of the work in front of it, inside the terminal it's running in. Run five sessions and you're switching between five tabs to see where each one stands. When someone asks "where are you on this?", you're back to scrolling chats or asking the agent, which costs tokens and interrupts its work.
Leftmark sits above all of them: one board, every session, every agent.
What's on the board
| Lane | What lands there |
|---|---|
| Needs you | Questions from agents, runs waiting for your go-ahead, and work an agent says is ready for you to check. Reply on the card without opening anything. |
| Doing | Work an agent is on right now, with its logo and a turning ring while it runs. |
| Asked | Requests that are captured but not started yet. |
| Done | Closed tickets, with what was done. |
Every ticket keeps the project, git branch and session that worked on it, so you can resume that exact Claude Code session later, even after the terminal is closed.
How Claude Code fills the board
- Connect once. The app adds Leftmark to Claude Code as an MCP server, after backing up your settings. Codex and Gemini connect the same way.
- Work as usual. Ask Claude Code for something and it opens a ticket for the request.
- The agent keeps it current. It moves the ticket as it works, notes where it left off, and says when it's blocked.
- You glance. Press U to step through only what changed while you were away.
A dashboard for every AI agent, not just Claude Code
Leftmark works as an AI agent dashboard across tools. Claude Code, Codex and Gemini write to the same board, each card showing which agent owns it. To hand work over, write @codex fix the flaky test on a ticket and Codex starts a run in its own git worktree. When one agent asks another for help, you approve it first.
It tracks. It doesn't orchestrate.
Tools like Vibe Kanban and Claude Squad run your agents for you, in their own interface. Leftmark doesn't ask you to change how you work. You keep using Claude Code in your terminal; the board fills itself. Compare: Leftmark vs Vibe Kanban · Leftmark vs Claude Squad.
Private by design
- The board is one SQLite file on your Mac. No account, no cloud copy.
- No telemetry, and no AI calls of its own.
- Projects claim folders, so an agent in your day job's repository can't read your side project's tickets.
Put your Claude Code sessions on one board
In beta for Apple Silicon Macs on macOS 13 or later. Invite only for now.
Request an invite →Questions
Is there a kanban board for Claude Code?
Yes. Leftmark is a Mac app, now in beta, that connects to Claude Code over MCP. Every request becomes a ticket that Claude Code moves from Doing to Done itself, sending it to Needs you when it wants you to check something.
Does it work with several Claude Code sessions at once?
Yes. Every session writes to the same board, and each ticket keeps the session, project and branch that worked on it.
Do I have to change how I use Claude Code?
No. You keep running Claude Code in your terminal. Leftmark adds itself as an MCP server, and the agent keeps the board up to date.
Does it work with Codex and Gemini too?
Yes. Codex and Gemini connect the same way, and their work shows on the same board with each agent's logo on its cards.