The kanban board for vibe coding
Vibe coding fails when the plan lives in a chat scrollback. Kanboard makes the plan a board: you break the idea into tasks, your agent works them over MCP, and you watch cards move instead of re-reading transcripts.
Sound familiar?
The plan is buried in chat history
Yesterday's session ended mid-plan. Today's agent has no idea — the plan scrolled away with the context window.
Agents wander off-plan
Give an agent a vague goal and it invents scope. Without a task list to anchor it, 'build the app' becomes anything.
No sense of progress
After three hours of prompting, what's actually done? A wall of diffs isn't an answer.
How Kanboard fits
The board is the spec
Break the idea into tasks first. The agent reads the board over MCP and works the queue — scope is whatever's in To Do, nothing more.
Live progress
Cards move To Do → In Progress → Done while the agent works. Progress is a glance, not an interrogation.
Persistent memory between sessions
Tomorrow's session starts with get_board, not a recap. The board carries state across context windows.
One-command setup
claude mcp add with your API key and the connection is live — no plugins, nothing to host.
A week on the board
- 1
Describe the thing you want to build, then split it into 5–15 concrete tasks on the board.
- 2
Tell your agent: 'work the board — top of To Do first, one task at a time, move cards as you go.'
- 3
Interrupt only to reorder To Do when priorities change; the queue is the steering wheel.
- 4
When Done fills up, review, test, and ship — then load the next batch of tasks.
FAQ
What is vibe coding with a kanban board?+
Instead of steering an AI agent prompt-by-prompt, you put the plan on a board. The agent reads tasks over MCP, works them in order, and moves cards as it finishes — you supervise the queue, not the keystrokes.
Which agents can work the board?+
Claude Code, Codex, Cursor, Windsurf — any MCP client that supports remote Streamable HTTP servers with an Authorization header.
How do I stop the agent from doing too much?+
Small tasks with acceptance criteria in the description, and an instruction to work one card at a time. The board's scope is the agent's scope.
Do I need to know how to code?+
It helps — you'll review what agents ship. But the board itself is just tasks and columns; writing a good task is writing a good prompt.
A board for you, an API for your agents
One flat price for you and every agent you connect.
Start building