For coding agents
AI Agent Board
A hosted kanban your AI coding agents can read and update. Plan the work yourself; let Claude Code, Codex, or Cursor pull tasks over MCP.
What people mean by ai-agent-board
“AI agent board” and “ai-agent-board” usually mean the same thing: a visual queue for agentic coding, not a human-only project tool with an API bolted on later. Cards are the interface. Columns are state. Agents are first-class users of that state.
Some open-source boards run entirely on your laptop and start Claude Code (or Copilot, Codex, OpenCode) as a local process when you drag a card. That is useful if you want the board to orchestrate runs. Kanboard solves a different slice: the persistent, shared AI agent kanban (hosted, authenticated, reachable from any MCP client) while you still start Claude yourself in the terminal or your editor.
How Kanboard works as an AI agent board
Agents connect over MCP
The board is a remote MCP server. Claude Code and other clients call get_board, create_task, and move_task. You do not paste the backlog into every new session.
One queue, many agents
Convention: move a card to In Progress before starting. That is how two agents avoid grabbing the same work. The board is the lock file you can actually see.
You still own the backlog
An AI agent board is not an autopilot for product decisions. You break work into cards; agents execute. In Review is the human gate.
Keys you can revoke
Agents act through API keys from Settings → Integrations. Revoke a key and that client loses the board. MCP clients can also ask before each write.
A typical loop
- You capture work as small cards in To Do (one shippable slice per card, with acceptance criteria in the description).
- An agent (often Claude Code, Codex, or Cursor) reads the board, moves the top card to In Progress, and implements it.
- Finished work lands in In Review. You check the diff or PR, then move the card to Done, or send it back with notes.
Step-by-step agent workflow: kanban for AI agents. Multi-agent coordination: kanban for AI agent teams.
Kanban for Mac, same AI agent board
If you want a macOS kanban app in the dock, Kanboard ships a native Mac build. It is the same AI agent board you use in the browser, not a second copy of your tasks. Kanban for Mac covers the desktop app; download for macOS if you already have an account. There is no native Windows or Linux app yet.
Frequently Asked Questions
What is an AI agent board?+
An AI agent board is a kanban (or similar task board) that coding agents can read and update, not just a UI for humans. Cards are the work queue: agents pull a task, mark it in progress, and report done, so you see the fleet without grepping chat logs.
How is Kanboard different from a local ai-agent-board app?+
Many GitHub projects named along the lines of ai-agent-board run on your laptop and spawn Claude Code or Copilot as child processes. Kanboard is hosted: agents connect over MCP to a remote board. The queue is available from any machine, any session, and more than one agent at once. Kanboard does not launch agent processes for you.
Which agents can use this AI agent board?+
Any MCP client that supports remote Streamable HTTP with an Authorization header: Claude Code, Claude Desktop, Codex, Cursor, Windsurf, and others. One API key (or several, if you want to revoke them separately) is enough.
Can I use the AI agent board as a macOS kanban app?+
Yes. There is a native macOS app included with Pro, and the same boards run in the browser. Agents keep working the hosted board whether you are looking at the Mac app or a tab.
How much does an AI agent board on Kanboard cost?+
Pro is $12/month: unlimited projects and tasks, MCP, REST API, and the Mac app. Agents are not extra seats.