Quickstart
Once installed, you can drive Coolify with natural language. Here are five things to try.
1. Survey the infrastructure
"What servers and apps do I have?"
The model uses get_infrastructure_overview for a compact summary, or list_servers + list_applications for more detail.
2. Diagnose a sick app
"Why is
my-app-namereturning 500s?"
The model uses diagnose_app, which aggregates application config, recent deployments, logs, and env-var metadata into a single response. It can spot things like "deployment failed 12 minutes ago" or "healthcheck disabled."
3. Push a config change
"Add
LOG_LEVEL=debugto staging-api as a runtime-only variable."
The model uses env_vars with action: 'create', is_buildtime: false, is_runtime: true. The runtime-only flag matters: variables marked as build-time are injected as Dockerfile ARG, which can break multiline secrets like PEM keys.
4. Bulk operation
"Set
NODE_ENV=productionon every app in the prod environment."
The model uses bulk_env_update with the project's app UUIDs. Safer than scripting it manually because the response shows which apps succeeded and which failed.
5. Deploy
"Deploy the latest from main on the api app."
The model uses deploy with the application UUID. Currently blocks until the deployment completes. (The v3 Tasks primitive will return immediately with a task ID instead.)
Tools that are read-only (safe to let the model run freely)
- All
list_*andget_*tools diagnose_app,diagnose_server,find_issuessystem({ action: 'health' }),system({ action: 'list_resources' })get_infrastructure_overviewsearch_docs
Tools that change state (consider per-call confirmation in your client)
- All
create_*/update_*/delete_*actions deploy,redeploy_project,restart_*stop_all_apps(the emergency-stop button)bulk_env_update
The v2.x tools don't yet declare destructiveHint / readOnlyHint annotations to drive client confirmation UX. That's a v3 add (see v3 vision).
What next?
- Tools reference — every one of the 42 tools, grouped by concern, marked read or destructive.
- How coolify-mcp works — the architecture, the file structure, the security model.