Deployment

Docker Compose (recommended)

git clone https://github.com/t11z/claudecord.git
cd claudecord
docker compose up -d

Put DISCORD_BOT_TOKEN and DISCORD_APPLICATION_ID in a .env file first (see .env.example) — Compose picks it up automatically, and there’s no dashboard form for the bot token (see Getting started for why). Everything else — inviting the bot, signing into the dashboard, linking Claude/GitHub — happens after the bot is running.

Volumes — do not skip this

The compose file mounts two volumes. Both matter:

Volume	Contains	If you lose it
claudecord-data → /data	SQLite DB, secrets, per-thread workspaces	all config & thread mappings gone
claudecord-claude → /home/bot/.claude	Claude Code’s session storage	every conversation forgets its history on restart

Claude Code persists conversation state on disk, keyed by each session’s working directory. That’s why ~/.claude needs a volume and why the bot never changes a thread’s workspace path.

Exposing the dashboard

By default the dashboard is only reachable from the machine itself (127.0.0.1:3000). To reach it from elsewhere:

DASHBOARD_HOST=0.0.0.0
DASHBOARD_PUBLIC_URL=https://claudecord.example.com

There’s no password to set — every route requires a session, and a session can only come from redeeming a /dashboard magic link (see Dashboard accounts). DASHBOARD_PUBLIC_URL matters here specifically: it’s the base URL baked into that link, so it has to be whatever address actually reaches the dashboard from the browser you’ll sign in with — not localhost, once the bot and the browser aren’t on the same machine. Put a reverse proxy with TLS in front for anything beyond your LAN.

Bare Node

npm install
npm run build
cp .env.example .env   # fill in tokens
npm start

Node ≥ 20 required. Use a process manager (systemd, pm2) for anything long-lived, and back up ./data plus ~/.claude.

Updating

git pull
docker compose up -d --build

Database migrations run automatically on startup.

Host sizing

Each concurrent Claude run spawns a Claude Code subprocess (roughly a small Node process each). The default cap of 4 concurrent runs is comfortable on 1 vCPU / 1 GB. Raise MAX_CONCURRENT_RUNS only with the RAM to match.