Skip to main content
Get Started

v1.2.1 · Open source · MIT

Kanban boards
that live in your repo

Plain-markdown task management with a web UI, CLI, REST API, and MCP server for AI agents. No database, no SaaS, no lock-in.

Get started in 60s View on GitHub 
npm install -g kanban-lite
kl init && kl serve

Pick your interface — all four ship in the same package

Web UI

Drag-and-drop board, inline editor, real-time WebSocket sync. kl serve

See the tour →

CLI

Create, filter, move, and report from the terminal. kl list

CLI reference →

REST API

OpenAPI-documented JSON API with Swagger UI included. Automate anything.

API reference →

MCP Server

Let Claude, Codex, or any MCP-compatible agent manage your backlog.

MCP reference →

Up and running in three commands

Install

npm install -g kanban-lite

Init

kl init

Creates .kanban/ in your project.

Go

kl serve

Opens the board at localhost:3000.

Full quick start guide →

What you get out of the box

Markdown-native cards

Every card is a plain .md file with YAML frontmatter. Read, edit, and commit them like any source file. No hidden format.

Multi-board support

Run separate boards for features, bugs, sprints, or any context. Each board has independent columns, settings, and labels.

Rich card model

Priority, assignees, due dates, labels, attachments, comments, logs, card actions, and structured forms — all in plain files.

Powerful filtering

Fuzzy search, metadata token filters (meta.field: value), clickable label chips, priority and assignee filters in every interface.

Plugin ecosystem

Swap the storage backend (SQLite, MySQL, S3), add auth, or extend webhooks — all via .kanban.json configuration.

Git-friendly by design

Everything lives in your repo. Diffs, blame, pull request reviews, and rollbacks work on your board exactly as they do on code.

Who it’s for

Individual developers

Keep your task list next to your code. No extra tool, no browser tab, no account. Just kl add and commit.

Start solo →

Engineering teams

Shared board in the monorepo. Every status change is a commit. Sprint reviews happen in code review — no extra dashboards.

Team sprint example →

AI agents

Claude, Codex, and OpenCode can read your backlog, create cards, and move tasks through the workflow using the built-in MCP server.

Set up MCP →
  • Markdown-native storage
  • MIT licensed
  • No database required
  • VS Code extension included
  • Zero cloud dependency
  • Real-time WebSocket sync

Ready to try it?

Install the CLI, init a board, and be running in under a minute.

Get started Read the FAQ