agent-kanban
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@agent-kanbanshow me the kanban board"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
agent-kanban
A shared kanban for people and their agents. Tasks, hand-offs, reviews, notes and ideas live as files in a private GitHub repo. Everyone works on them from the agent they already use: Claude, ChatGPT, Claude Code, Codex, or anything that speaks MCP.
Part of warOnSaaS. Licensed AGPL-3.0. UI from the warOnSaaS ui-design kit (lib/ui/kit.css, synced from ui-design main with node scripts/sync-kit.mjs). A team restyles its board with brand/brand.json (see lib/brand.mjs).
Two ways to run it:
Create your board. Sign in with GitHub, name your team, done: agent-kanban-hosted.vercel.app (a preview for now). We run the app; your board's data goes into a private repo in your own GitHub account.
Host it yourself, free. Your server, your GitHub repo, no account with us. The steps.
Either way, your data is always in your own GitHub repo, and a board we host can move to your own hosting any time with one command (npx -y github:warOnSaaS/agent-kanban deploy).
WHAT IT DOES
One board per team. Clients, tasks, notes, docs, ideas, alerts. Plain markdown files in your repo, with history.
Sign in with GitHub. Who you are comes from GitHub; what you can see comes from
people.yml.Access per person: owner sees everything; teammates see their clients;
sees: ownlimits someone to their own tasks; any item can be private to named people. Enforced by the server: the repo itself goes to the owner only, because GitHub can't hide folders.Hand-offs: "what I did / what's next" goes into the task, the next person's agent picks it up.
Review: the agent opens the actual work (linked pages, linked files), says Ready or Not ready, up to 3 issues, then approve / send back / meet.
Alerts on hand-offs, comments and changes. Unread count on every reply.
Short commands:
start,check tasks,review,new task,assign task,hand off task,status,view kanban.view kanbandraws the board inside Claude and ChatGPT chats (MCP Apps);/boardshows it in a browser, by client.
Related MCP server: KittyClaw
HOW PEOPLE JOIN
Send them your board's address. Its front page is a row of app tiles: they pick their app, click once, sign in with GitHub, and type start.
Tile | What the click does |
Claude (web, desktop, phone) | Opens Claude's "Add custom connector" form already filled in. They click Add, then Connect. |
ChatGPT | Opens your team's GPT (shown only when you set |
Claude Code, Codex | Shows one line to paste in a terminal: |
Browser | Opens the board. |
Agents can still read /INSTRUCTIONS.md, which walks any agent through joining. People never need to read it.
HOST IT YOURSELF
About 20 minutes. You need a GitHub account and a free Vercel account. Nothing to install.
Make the private repo that holds your board. Go to github.com/new, name it (for example
acme-board), choose Private, and create it. Then copy theexample-workspacefolder from this repo into it: on your new repo's page click Add file, Upload files, and drag in everything insideexample-workspace.Put your team in
people.yml. In your new repo, openpeople.yml, click the pencil, and replace the made-up people with your own. For each person: a short id, their name, their GitHub username,role: ownerfor you androle: teamfor everyone else, andclients:all, or the client folders they work on. Save. Theclientsfolder holds two made-up clients to show the shape. Keep them while you try it, then rename or delete them.Put the app online. Open Deploy to Vercel. Vercel copies this app into your GitHub and asks for five settings; the table below says what to put in each. When it finishes, Vercel shows your board's address, for example
https://acme-board.vercel.app. That address is what you send people.Turn on "Sign in with GitHub". Go to github.com/settings/developers, click New OAuth App, and fill in:
Application name: your team's name
Homepage URL: your board's address from step 3
Authorization callback URL: your board's address followed by
/oauth/github/callback, for examplehttps://acme-board.vercel.app/oauth/github/callback
Click Register, then Generate a new client secret. Put the Client ID and the secret into Vercel (your project, Settings, Environment Variables) as
GITHUB_OAUTH_CLIENT_IDandGITHUB_OAUTH_CLIENT_SECRET, then redeploy (Deployments, the three dots on the latest one, Redeploy).Send people your board's address. They pick their app and type
start.
The settings
Setting | What to put |
| Your repo from step 1, written as |
| A key that lets the app read and write that repo. Make it at github.com/settings/personal-access-tokens/new: pick only that repo, and give Contents and Administration read and write. |
| Any long random text. It signs people's sessions. Keep it secret. |
| Your team's name as people see it, for example |
| Your time zone, for example |
| Who people message when they get stuck, for example |
| From step 4 |
Optional:
Setting | What it does |
| That person's email, offered when someone's GitHub account uses a different email |
| Your team's ChatGPT GPT link. Shows the ChatGPT tile. How to make the GPT: |
| Needed for that GPT, see the same doc |
| Emails urgent alerts |
| Instead of |
| Your board's https address, when you run |
|
|
Your board can look like your brand: add brand/brand.json and a logo to your repo (see lib/brand.mjs).
HOSTED BOARDS, AND MOVING ONE
One deployment can also serve many teams (HOSTED=1, deployed with vercel.hosted.json). Each team's board is at <host>/t/<team>/, its MCP address at <host>/t/<team>/mcp, and its data in a private repo in the team owner's own GitHub account. <host>/mcp lets an agent make and manage boards: "make me a board for my team". How it works, the isolation rules and the setup: docs/HOSTED.md.
To move a hosted board to your own Vercel account (your repo stays where it is, people are sent to the new address):
npx -y github:warOnSaaS/agent-kanban deploy --repo you/your-board --from https://<host>/t/<team>Add --dry-run to see every step first. --target node prints the steps for any other Node host. The owner can also download everything as a zip at any time (/export.zip, or ask your AI app for the link).
WORKS WITH
Claude and Claude Code are trademarks of Anthropic. ChatGPT, Codex and the OpenAI logo are trademarks of OpenAI. The connect page shows their marks, unmodified, only to say which apps the board works with. agent-kanban is not made or endorsed by Anthropic or OpenAI.
DEVELOP
npm install
npm test # end-to-end tests against example-workspace and a fake GitHub
npm run dev # local board on example-workspace at http://localhost:3977
DEMO_BOARD=1 npm run dev # the demo board, no sign-in
HOSTED=1 npm run dev # many teams, "Create a board" (a preview with no GitHub App: made-up boards)
npm run check:clean # no instance names in tracked filesRead AGENTS.md before changing anything: every change is TEMPLATE, INSTANCE CONFIG or INSTANCE CONTENT.
This server cannot be deployed
Maintenance
Related MCP Connectors
Task & board management for AI agents + humans. Kanban, comments, digests via MCP.
AI-native Kanban board — connect Claude to claim, work and move your tasks over MCP.
Kanban board for teams and coding agents: manage tasks, subtasks, sprints and wiki pages via MCP.
Remote MCP for Kanban AI boards—manage projects, tasks, and comments from AI tools.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceA local kanban board and docs tool that AI agents can drive over MCP. It enables project management without accounts or API keys.MIT
- AlicenseNot gradedqualityAmaintenanceSelf-hosted kanban board that dispatches AI agent fleets against tickets. The embedded Streamable HTTP /mcp endpoint exposes 7 tools to list projects, read boards and tickets, and create, move and comment tickets from any MCP client.27AGPL 3.0
- AlicenseNot gradedqualityAmaintenanceEnables AI agents and humans to collaboratively manage kanban boards and Markdown documentation via MCP tools, with stable item keys, revision-safe editing, and full audit trails.MIT

AgentTaskerofficial
AlicenseNot gradedqualityBmaintenanceEnables AI agents to create, query, update, claim, hand off, and complete tasks on a local SQLite-backed, project-namespaced kanban board through stdio MCP tools, including filtering for ready/blocked/stale work and exchanging task attachments. It also lets agents list projects and manage statuses, ownership, and evidence without any server or accounts.MIT