Skip to main content
Glama

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 wOS UI kit (lib/ui/wos.css, synced from warOnSaaS/site with node scripts/sync-kit.mjs).

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: own limits 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 kanban draws the board inside Claude and ChatGPT chats (MCP Apps); /board shows it in a browser, by client.

  • Agents act as chief of staff: one line per item, detail on request.

Related MCP server: KittyClaw

HOW PEOPLE JOIN

Send them https://<your-instance>/INSTRUCTIONS.md. They hand it to their agent. The agent works out what it is (Claude Code, Codex, Claude app, ChatGPT) and walks them through: GitHub account, connect, sign in, start. The file holds no secrets.

RUN YOUR OWN INSTANCE

  1. Workspace repo. Create a private GitHub repo. Copy example-workspace/ into it. Edit people.yml: GitHub usernames, roles, clients.

  2. GitHub OAuth app. github.com/settings/developers, New OAuth App. Callback URL: https://<your-instance>/oauth/github/callback.

  3. Deploy this repo to Vercel (or anything that runs Node functions; dev.mjs is a plain Node server). Environment:

Variable

What

WORKSPACE_REPO

owner/repo of the workspace repo

GITHUB_TOKEN

Token that can read and write that repo (fine-grained: Contents read/write, Administration read/write to invite the owner)

GITHUB_OAUTH_CLIENT_ID, GITHUB_OAUTH_CLIENT_SECRET

From step 2

OAUTH_SECRET

Long random string; signs sessions and tokens

WORKSPACE_NAME

Shown to people, e.g. Acme Ops

WORKSPACE_TZ

e.g. America/New_York (default UTC)

WORKSPACE_CONTACT

Who people message when stuck

WORKSPACE_CONTACT_EMAIL

Their email, offered when someone's GitHub account uses a different email

OAUTH_CLIENT_ID, OAUTH_CLIENT_SECRET

Optional, for the ChatGPT GPT. See docs/chatgpt-gpt.md

RESEND_API_KEY

Optional, emails urgent alerts

  1. Send people INSTRUCTIONS.md.

WHERE IT WORKS

App

How

Claude Code

claude mcp add --transport http --scope user agent-kanban https://<instance>/mcp, then /mcp to authenticate

Codex

codex mcp add agent-kanban --url https://<instance>/mcp, then codex mcp login agent-kanban

Claude (web, desktop, phone)

Settings, Connectors, Add custom connector, URL https://<instance>/mcp

ChatGPT (any paid plan, web and phone)

A GPT with Actions from /openapi.json. See docs/chatgpt-gpt.md

Browser

https://<instance>/board, read-only

DEVELOP

npm install
npm test                 # 23 end-to-end tests against example-workspace and a fake GitHub
npm run dev              # local server on example-workspace
npm run check:clean      # no instance names in tracked files

Read AGENTS.md before changing anything: every change is TEMPLATE, INSTANCE CONFIG or INSTANCE CONTENT.

Maintenance

ActivityMaintained
ResponsivenessNo issues

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    A
    maintenance
    Self-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.
    27
    AGPL 3.0
  • A
    license
    Not graded
    quality
    A
    maintenance
    Enables 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
  • A
    license
    Not graded
    quality
    B
    maintenance
    Enables 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