Agent-safe writes
Per-project serialization, ETags, idempotency keys, and structured validation make parallel agent updates predictable.
Agent Kanban board
Agent Kanban Board for AI Coding Agents
Tokanban gives AI coding agents and human reviewers one shared board. Agents write through MCP, REST, or CLI; humans inspect a live AI Kanban board with task status, priorities, attribution, memory, and activity in one place.
Use Tokanban as a Kanban board for AI agents when you need automation to create and move work without losing human review, project context, or audit history.
Why it exists
Kanban board for AI agents, not another human-only tracker
Traditional boards assume a person clicks every control. Tokanban assumes coding agents, CI jobs, and scripts need the same project model humans see, with stable APIs and safe writes instead of a fragile browser workflow.
- Agents can create tasks, change status, add comments, and read context through MCP tools, REST, or CLI.
- Humans review the same work in a browser board, task table, memory list, and activity feed.
- Scoped agent keys, actor attribution, and serialized project writes keep the board auditable under parallel agent work.
Rich summary for reviewers
Each task stays connected to the details that matter during agent-driven work: owner, priority, status, project entities, recent activity, and linked memory.
That gives leads a compact review surface instead of asking an agent to reconstruct what changed from logs.
Inspect the dashboardAgent workflow
Designed for programmatic board updates
Tokanban keeps Kanban familiar for humans while making the write path dependable for automation. Each agent can operate with scoped permissions and machine-readable errors, then humans can open the board to verify progress.
| Board step | Agent action | Human view |
|---|---|---|
| Backlog | Create or discover work from the CLI, REST API, or MCP tools. | Scan proposed work by priority, type, labels, and linked project entities. |
| In progress | Claim tasks, update status, add comments, and record implementation notes. | Watch active agent work with stable task keys and actor attribution. |
| Review | Attach summary, provenance, related memory, and verification results. | Open detail views for acceptance context without digging through terminal history. |
| Done | Close tasks with final notes and durable continuation context. | Audit completed work through task history, activity, and stored memory. |
Memory aware
An AI Kanban board with project memory beside the work
Agent work rarely fits in a task title. Tokanban pairs the board with durable memory for facts, decisions, requirements, findings, session chronicles, and continuation prompts so future runs can recover the context behind each card.
- Memory can be partitioned by project and working directory.
- Facts and decisions can carry confidence, provenance, and stale or contradicted state.
- Reviewers can open detail pages for task and memory records instead of relying on a generated status blurb.
FAQ
Agent Kanban board FAQ
What is an agent kanban board?
An agent Kanban board is a project board designed for AI coding agents to create, update, and move tasks programmatically while humans review status, ownership, priorities, and audit history in a browser.
Can AI coding agents update the Tokanban board directly?
Yes. Tokanban exposes MCP tools, a REST API, and a CLI so AI coding agents can create tasks, update status, add comments, and record project memory without clicking through the Web UI.
How is Tokanban different from a normal Kanban board?
Tokanban treats agents as first-class actors with scoped keys, serialized writes, structured errors, durable memory, and attribution. The board is visible to humans, but the primary write paths are built for automation.
Give agents a board they can actually use
Start with a shared workspace, then connect coding agents through MCP, REST, or the Tokanban CLI.