Appearance
Coolify
Coolify v4 runs everything on the server: builds, containers, the Traefik proxy, databases, backups and scheduled tasks. It lives at coolify.stu-bot.uk, behind Cloudflare Access.
Settings that matter
- Registration: off. The first account becomes root, and it has to be created within 10 minutes of Coolify starting. Restarting the
coolifycontainer reopens that window. - API: on, at
/api/v1, with bearer tokens. - MCP server: on, at
/mcp. This is what Claude Code talks to. It has about 80 tools covering apps, deployments, logs, env vars, databases, backups, DNS, services and server metrics. - Sentinel: on, sending server metrics (CPU, memory, disk, load, network) with 7 days of history.
- Server: "localhost". Coolify reaches the host over SSH as root from its own container, which the sshd settings allow from the Docker networks only.
Tokens
| Token | Permissions | Used by |
|---|---|---|
| Claude Code (Mac) | root | the infra plugin and CLI |
| ops-service | read, deploy | the ops page (restart and redeploy buttons) |
| claude-runner-readonly | read | Claude on the server |
All of them expire after a year. Resetting the root password revokes every one of them.
GitHub
Coolify has its own GitHub App, installed on all my repos. On push, GitHub calls https://coolify.stu-bot.uk/webhooks/... (an Access bypass path), and Coolify deploys whichever apps watch the changed paths.
If you register a GitHub App while logged out of Coolify, the GitHub callback is lost. Log in first.
Projects
Resources are grouped into Coolify projects, each with a production environment. The shared pieces of the stack live in a project called platform:
| Resource | Type | Reach |
|---|---|---|
| bus | Valkey 9 | internal |
| n8n | service: n8n + Postgres + task runners | n8n.stu-bot.uk |
| ops | app from apps/ops | ops.stu-bot.uk |
| claude-runner | app from apps/claude-runner | internal |
| docs | app from apps/docs | docs.stu-bot.uk (public) |
Notifications
Coolify sends its own Telegram messages for failed deploys, failed backups, failed scheduled tasks, disk usage, the server becoming unreachable (and reachable again), and Traefik updates.
Things that caught me out
- An app created without a domain gets an automatic public
*.sslip.ioaddress. For internal apps, set the domain to empty and checkfqdncomes back null. The firewall now blocks direct traffic anyway, but don't rely on that alone. - Coolify's health checks run
curl(falling back towget) inside the container, so slim images need one of them installed. - A database inside a Coolify service, like n8n's Postgres, can't have backups created through
/databases/{uuid}/backupsin the API. Use the service's Backups tab.