ClickDown

A fast, lightweight, multi-team Kanban task manager built on Bun and SQLite. ClickDown uses a hypermedia-driven architecture — server-side rendering with HTMX and Alpine.js — to deliver a responsive, app-like experience without a heavy client-side framework or build step.

It is local-first: a single Bun process provides the database (embedded SQLite), file storage (data/uploads/), and real-time updates (in-process WebSockets). No CDN, cloud service, or external broker is required — the app runs fully offline or on a LAN.

✨ Features

Tasks

• Title, description (with Markdown rendering + Write/Preview toggle), priority (low / medium / high / urgent), assignee, and due date

• Inline editing of every field from a task detail modal

• Checklists, comments, labels, and file attachments (stored under data/uploads/)

• Duplicate a task, or bulk-select and delete many at once

• Archive tasks (soft-hide) with a recycle-bin style archive view; permanent purge (single task or empty archive)

• Activity log per task tracking who changed what

• Shareable deep links — every task has a /tasks/:id URL that opens its board with the task modal expanded

• Priority color-coding and overdue / due-soon badges

• Dashboard summary of assigned tasks with team/board context

Boards

• Boards created with a default Backlog → To Do → In Progress → Done flow

• Customizable columns (add, edit, delete, reorder, color, WIP limits) — admin/owner only

• Drag-and-drop tasks within a column, across columns; column reorder for admins

• Live task search across the board, "My tasks only" filter, and "Show archived"

• Keyboard shortcuts (/ focuses search, n adds a task, Esc closes modals)

Teams & collaboration

• Create teams, or join via a tokenized invite link (slug alone is not enough; owners/admins can regenerate the link)

• Owner / admin / member roles, with a member-management UI to change roles and remove members

• Isolated boards and tasks per team with membership-based access control

• Board/column structure mutations require admin or owner; task work is open to all members

Real-time (WebSocket)

• Live presence — avatars appear/disappear as team members open or close the board; new viewers receive a roster of who's already there

• Live task updates — create, edit, move, delete, and archive operations broadcast to all viewers instantly

• Refresh safety — remote changes reload the board with your search/filters preserved, and are deferred (with a notice) while you have a dialog open so unsaved edits are never lost

• Built on Bun's native server.publish() + HTMX WS Extension + Alpine.js; no external message broker required

Accounts & UX

• Registration / login / logout with bcrypt-hashed passwords (cost 12) and strength validation (8+ chars with upper, lower, digit, and special character)

• Profile settings to update display name, email, and password (email or password change requires your current password; other sessions are revoked)

• API tokens for AI agents — create/revoke personal tokens from the profile page (see MCP below)

• Auto-generated initials avatars

• Light / dark theme toggle (persisted to localStorage)

• Toast notifications for all actions — visible even above open modals, and on network failures

• Responsive layout and a custom design system over a modernized Pico CSS base

• Structured logging via pino; health probes at /health and /ready

AI agent API (MCP)

• Model Context Protocol endpoint at POST /api/mcp (Streamable HTTP, JSON-RPC 2.0)

• Personal bearer tokens (SHA-256 hashed at rest) created from the profile page, with a one-step claude mcp add command shown on creation

• 16 tools covering teams, boards, tasks, checklists, comments, and search — mutations broadcast live to open boards

Related MCP server: Planka MCP Server

🛠 Tech Stack

🚀 Getting Started

Prerequisites

Bun 1.3.14 is the only requirement. Install or upgrade it if needed:

curl -fsSL https://bun.sh/install | bash
bun upgrade

If you use mise, the pinned mise.toml provides the same version automatically (mise install).

Installation

git clone <repository-url>
cd ClickDown
bun install

Database Setup

bun run dev and bun run start apply pending SQLite migrations automatically. You can also run them explicitly:

bun run db:migrate

By default the database lives at data/clickdown.db (override with DB_PATH).

Running the App

bun run dev     # development, with hot reload (recommended)
bun run start   # production

The app is then available at http://localhost:3000.

Health checks: GET /health (liveness) and GET /ready (schema and writable local upload storage).

All browser runtime assets are served locally from installed packages. Running ClickDown does not require CDN or cloud-service access.

Docker

A production-ready Dockerfile is included:

docker build -t clickdown .
docker run -p 3000:3000 -v clickdown-data:/app/data -e NODE_ENV=production clickdown

The container:

• Applies pending migrations before accepting traffic, and ships a HEALTHCHECK on /ready

• Installs production dependencies only in the runtime stage (no Playwright/TypeScript/Biome)

• Runs as the non-root bun user with the base image pinned by digest

• Persists SQLite data (and uploads) in a named volume (clickdown-data)

• Sets NODE_ENV=production (strict CSRF Origin checks)

• Declares STOPSIGNAL SIGTERM — the app shuts down gracefully (stops accepting requests, closes WebSocket connections, closes the database)

• Exposes port 3000

The SQLite, in-memory rate limit, and WebSocket design supports one application process. Do not run multiple replicas against the same database volume.

.dockerignore excludes test files, local DBs, and dev tooling for a minimal image.

Optional smoke: bun run smoke:docker — builds the image, polls /ready, checks /health, and always cleans up (requires Docker).

⚙️ Configuration

ClickDown is configured through environment variables (Bun auto-loads .env):

🤖 MCP (AI agent) endpoint

ClickDown exposes a Model Context Protocol endpoint so AI agents (Claude Code, Cursor, custom scripts, etc.) can find and work on tasks directly.

Endpoint: POST /api/mcp (Streamable HTTP transport, JSON-RPC 2.0)

Setup

1. Open Profile → API Tokens (MCP) and create a token. It is shown once — copy it.

2. Register ClickDown with Claude Code (the profile page shows this command pre-filled with your token and URL):

   claude mcp add --transport http clickdown http://localhost:3000/api/mcp \
     --header "Authorization: Bearer cd_your_token_here"

3. In Claude Code, run /mcp to verify the connection.

For config-file based clients (Cursor, etc.):

{
  "mcpServers": {
    "clickdown": {
      "type": "http",
      "url": "http://localhost:3000/api/mcp",
      "headers": { "Authorization": "Bearer cd_your_token_here" }
    }
  }
}

Tokens act as your user (full team access) and can be revoked at any time from the profile page. Only the SHA-256 hash is stored; session cookies are not accepted on the endpoint.

Available tools

All mutations broadcast over the existing WebSocket layer, so open boards update live when an agent makes a change. A typical agent flow: list_teams → list_boards → list_tasks → get_task → move_task + add_comment.

📁 Project Structure

src/
├── index.tsx             # Entry point: handleRequest() + Bun.serve()
├── config/               # Database + migrations runner
├── middleware/           # Session auth
├── mcp/                  # MCP protocol (JSON-RPC) + agent tool definitions
├── models/               # TypeScript types
├── realtime/             # WebSocket pub/sub + presence
├── routes/               # HTTP route handlers
├── services/             # Business logic
├── schemas/              # Zod validation
├── utils/                # access helpers, logger, password, HTML safety
└── components/           # Kita JSX pages, fragments, modals
public/                   # Static assets (custom.css, app.js, favicon)
migrations/               # SQL migrations (001–018, checksummed + auto-applied)
scripts/                  # migrate.ts, e2e-reset.ts
tests/                    # unit, integration, e2e
.github/workflows/        # CI: static gates, E2E, Docker smoke
mise.toml                 # Bun 1.3.14 toolchain pin
data/                     # SQLite DB + uploads (gitignored)

Request flow: Browser → HTMX → handleRequest → auth/access → service → SQLite → Kita JSX → HTML fragment.

🔒 Security notes

• XSS prevention is enforced in CI — @kitajs/ts-html-plugin (bun run xss:scan) fails the build on unescaped JSX; user content is escaped at render, and Markdown is sanitized with http(s)-only links

• Local-only CSP — all scripts/styles/fonts load from the same origin (vendored, no CDN), with default-src 'self'

• CSRF — Origin/Referer must match on every mutating request (required in production; APP_URL defines the external origin behind a proxy). The MCP endpoint is exempt: bearer tokens are never sent ambiently by browsers, and session cookies are rejected there

• Sessions — HttpOnly + SameSite=Lax cookies, 7-day expiry; password and email changes require the current password and revoke other sessions (including live WebSockets)

• API tokens — shown once at creation, stored as SHA-256 hashes, revocable instantly, deleted with the owning user; a periodic sweep closes WebSockets whose sessions or team access have been revoked

• Invites — secret, regenerable tokens; team slugs alone never grant access

• Rate limits — in-memory, scoped per endpoint and per user for auth, join, profile, token creation, markdown preview, uploads (10 MB per file), and task/board/team mutations

• Attachments — served only to team members, forced Content-Disposition: attachment

💾 SQLite backups

The database is a single file (default data/clickdown.db, plus -wal/-shm in WAL mode).

# Safe online backup while the app is running (SQLite CLI)
sqlite3 data/clickdown.db ".backup 'data/clickdown-backup.db'"

# Or stop the app and copy the files
cp data/clickdown.db data/clickdown.db-wal data/clickdown.db-shm /path/to/backup/

Also back up data/uploads/ if you use attachments. Prefer volume snapshots when running under Docker.

🧪 Testing

• Unit (tests/unit) — services/utils against in-memory SQLite

• Integration (tests/integration) — real handleRequest in-process, including the MCP endpoint

• E2E (tests/e2e) — Playwright against a server on port 3001, including security/hardening regressions (XSS, offline, realtime) and the MCP token flow. Runs on both Chromium and Firefox (two suites skip on Firefox: JS coverage and clipboard permission, which are Chromium-only Playwright APIs).

Install the browser binaries once before running E2E tests:

bunx playwright install chromium firefox

bun run test              # unit + integration
bun run test:coverage
bun run test:e2e          # full Playwright suite
bun run test:e2e:smoke    # smaller auth + security subset
bun run xss:scan          # JSX escaping audit (@kitajs/ts-html-plugin)
bun run ci                # typecheck + XSS scan + lint + unit/integration
bun run ci:full           # ci + full e2e

GitHub Actions (.github/workflows/ci.yml) runs the full gate set on push/PR: static checks + unit/integration, the Playwright suite, and a Docker build with a readiness-probed smoke test.

📋 Available Commands

📄 License

MIT © 2026 amscotti — free to use, modify, and self-host, with attribution.