Discord Furlow MCP Server

FURLOW

Declarative Discord Bot Framework

Build powerful Discord bots with YAML. No code required.

commands:
  - name: ping
    description: Check bot latency
    actions:
      - reply:
          content: "Pong! ${client.ws.ping}ms"

Features

• 85 Actions: messages, moderation, voice, channels, and more
• 71 Expression Functions: date, math, string, array manipulation
• 50 Transforms: pipe-based data transformations
• 59 Events across Discord gateway, voice transitions, component interactions, and high-level FURLOW events (automod, scheduler, pipes)
• Scoped State: global, guild, channel, user, and member scopes
• Flows: reusable action sequences with parameters
• External Integrations: HTTP, WebSocket, MQTT, TCP/UDP, webhooks

Quick Start

# Install the CLI
npm install -g @furlow/cli

# Create a bot
furlow init my-bot
cd my-bot

# Add your Discord credentials to .env
echo "DISCORD_TOKEN=your_token_here" > .env
echo "DISCORD_CLIENT_ID=your_client_id_here" >> .env

# Start
furlow start

Example

version: "0.1"

identity:
  name: "My Bot"

presence:
  status: online
  activity:
    type: playing
    text: "with FURLOW"

commands:
  - name: hello
    description: Greet someone
    options:
      - name: user
        type: user
        description: Who to greet
    actions:
      - reply:
          content: "Hello, ${options.user.display_name || options.user.username}!"

  - name: roll
    description: Roll a dice
    options:
      - name: sides
        type: integer
        description: Number of sides
        min_value: 2
        max_value: 100
    actions:
      - reply:
          content: "🎲 You rolled **${random(1, options.sides || 6)}**"

events:
  - event: member_join
    actions:
      - send_message:
          channel: "${env.WELCOME_CHANNEL}"
          content: "Welcome ${member.displayName}!"

CLI Commands

Documentation

Packages

Production deployment

A short checklist for shipping a bot to production. Each item lines up with a runtime guarantee — skipping one usually shows up later as silent data loss, an unscheduled feature, or a publicly forgeable session.

Required environment

# Discord
DISCORD_TOKEN=...
DISCORD_CLIENT_ID=...

# Storage (pick one)
DATABASE_URL=postgres://user:pass@host:5432/furlow   # production
# or for single-server bots:
SQLITE_PATH=/var/lib/furlow/bot.db

# Dashboard (only if running the web UI)
DASHBOARD_SECRET=$(openssl rand -hex 32)             # required in NODE_ENV=production
DISCORD_CLIENT_SECRET=...
DISCORD_CALLBACK_URL=https://dashboard.example.com/auth/discord/callback

The dashboard refuses to start in NODE_ENV=production without DASHBOARD_SECRET — there is no fallback default. WebSocket connections also require an authenticated session; unauthenticated clients are rejected at upgrade time and cannot subscribe to guild streams.

Storage

• Use Postgres for any deployment that runs more than one process or needs point-in-time recovery. The SQLite adapter is fine for single-server bots and tests, but write contention degrades quickly past a few thousand active users.
• Run database migrations on deploy. The SQLite adapter creates tables on demand; Postgres deployments should pre-create the tables defined in your state.tables: block.
• Back up the database. Builtins persist real state (warnings, levels, AFK status, ticket transcripts); none of this is reconstructible.

Scheduling and voice

• The CLI starts a CronScheduler that emits scheduler_tick every 60s. Builtins that rely on it (giveaways, polls, reminders) need the scheduler running — there is no separate cron process.
• Voice playback requires the @discordjs/voice peer plus ffmpeg on the PATH. The voice_track_start / voice_track_end events used by the music builtin only fire when the voice manager is initialized.

Process supervision

• Run furlow start under a process supervisor (systemd, pm2, Kubernetes Deployment) that restarts on exit. The bot does not self-daemonize.
• Forward SIGTERM so the gateway connection closes cleanly. The CLI flushes pending state on shutdown; killing with SIGKILL can lose the last batched writes.

Dashboard hardening

• Terminate TLS in front of the dashboard (nginx, Caddy, a load balancer). The session cookie sets secure: true in production and will not be delivered over plain HTTP.
• Restrict the OAuth callback URL in the Discord developer portal to the exact DISCORD_CALLBACK_URL you ship.
• POST endpoints whitelist their accepted fields and reject unknown or wrong-typed keys with 400. If you extend a settings shape, update the whitelist in apps/dashboard/server/routes/api.ts.

Observability

• /health returns process uptime; wire it to your liveness probe.
• /metrics exposes Prometheus-format counters. Add a scrape job pointed at the dashboard process.
• The CLI honors --verbose for structured per-event logging during incident triage.

License

MIT