The kanban board for AI agent teams
When several agents work one codebase, the bottleneck isn't the agents — it's coordination. Kanboard gives your agent fleet a shared task queue over MCP: one board as the source of truth for what's claimed, in flight, and done.
Sound familiar?
Agents lose the plot between sessions
Each session starts from zero. Without a persistent task source, you re-brief agents on state they should have been able to read.
No shared queue means collisions
Two agents grab the same problem, or a task falls between them. Chat threads don't arbitrate ownership — a board does.
Reviewing agent output is chaos
Work lands as a pile of diffs with no map back to intent. You need to see which task produced which change.
How Kanboard fits
MCP task queue
Agents call get_board, claim the top To Do task, and move it through columns — claiming is just move_task to In Progress.
Column state as progress log
Glance at In Progress to see what the fleet is doing right now; Done is your review queue, in order.
API keys per workflow
Create separate keys in Settings → Integrations so you can revoke one pipeline without stopping the rest.
REST API
Build custom dispatchers or dashboards on the same primitives the MCP server uses — projects, columns, tasks, moves.
A week on the board
- 1
Decompose the feature into small, independent tasks — each one completable by a single agent in a single session.
- 2
Add an AGENTS.md rule: check the board, take the top To Do task, move it to In Progress before starting.
- 3
Dispatch agents in parallel; the board arbitrates who owns what.
- 4
Review Done column tasks against their diffs; move anything that fails review back to To Do with notes in the description.
FAQ
How do multiple agents avoid working the same task?+
Convention: an agent moves a task to In Progress before touching code. Since move_task is available over MCP, put that rule in your AGENTS.md and agents follow it.
Which agents work with Kanboard?+
Any MCP client that supports remote Streamable HTTP servers — Claude Code, Codex, Cursor, Windsurf, and more. Point them at the MCP endpoint with an API key.
Can agents create their own tasks?+
Yes — create_task is part of the MCP toolset, so agents can file newly discovered work back into To Do instead of doing it silently.
Is there an audit trail of what agents did?+
The board reflects current state — column position and task descriptions. For deeper logging, mirror task events into your own store with the REST API.
A board for you, an API for your agents
One flat price for you and every agent you connect.
Start building