Coolify Orchestrator

Coolify Orchestrator

A stack-aware, MCP-optional skill that orchestrates your Coolify infrastructure.

First: Determine Operating Mode

Understand which mode you are operating in at the beginning of each session:

MCP_AVAILABLE = Can coolify-mcp MCP tools be called?
  Yes → MCP mode: use tools directly (references/api.md is unnecessary)
  No → API mode: use Coolify REST API via curl/bash (read references/api.md)

Even in MCP mode, the decision trees and playbooks in this SKILL.md are valid — MCP is just the transport layer.

Find or ask for environment variables:

• COOLIFY_BASE_URL — e.g. https://coolify.example.com
• COOLIFY_ACCESS_TOKEN — Coolify Settings → API Tokens

If these aren't in the env, ask the user. Do not ask for the second time — save it for the duration of the conversation.

---

Scenario Detection → Playbook Routing

Match the user's request with this tree:

Is it about "deploy" or "build"?
  + Turborepo / monorepo / package error → [TURBO-DEPLOY]
  + Normal app deploy → [DEPLOY-VERIFY]

Is it about "down" / "unhealthy" / "error"?
  + Container/file issue (htaccess, config, .env) → [CONTAINER-EXEC]
  + API/service issue → [DEPLOY-VERIFY] (diagnose flow)

Is it about "new" / "provisioning" / "tenant"?
  → [PROVISIONING]

Is it about "env var" / "environment" / "secret"?
  → [MULTI-TENANT-ENV]

Is SSH / container terminal / exec requested?
  → [CONTAINER-EXEC]

Read the relevant reference file for each playbook.

---

[TURBO-DEPLOY] Turborepo + Coolify Deploy Playbook

> Detail: references/turborepo-deploy.md

When to trigger: When a Turborepo monorepo is deployed on Coolify and the following errors are seen:

• Cannot find module '@myapp/shared'
• tsconfig.json cannot be found
• Shared package acts as if it hasn't been built
• TypeScript error in the middle of Docker build

Quick diagnosis:

1. Check the COPY order in the Dockerfile — shared packages must be built first
2. Is there a dependency pipeline in turbo.json?
3. Is Coolify's Docker build context correct? (monorepo root or subdirectory?)

# API mode — fetch build logs
curl -s -H "Authorization: Bearer $COOLIFY_ACCESS_TOKEN" \
  "$COOLIFY_BASE_URL/api/v1/applications/{uuid}/logs" | tail -100

In MCP mode: Use application_logs(uuid) → deployment(action: list_for_app) to find the last deploy.

When you identify the problem, read references/turborepo-deploy.md — fix patterns are there.

---

[DEPLOY-VERIFY] Deploy → Automatically Verify Loop

> Detail: references/deploy-verify.md

Objective: Trigger deploy, verify healthy operation, diagnose if there is an error.

Steps:

1. Trigger deploy
   MCP: deploy(uuid, force_rebuild?)
   API: POST /api/v1/applications/{uuid}/start

2. Monitor deploy status (max 3 minutes, 10s intervals)
   MCP: deployment(action: get, uuid: deploy_uuid)
   API: GET /api/v1/deployments/{uuid}
   Expected: "finished" — if "failed", fetch logs

3. Application health check
   MCP: diagnose_app(uuid or domain)
   API: GET /api/v1/applications/{uuid} → status: "running:healthy"?

4. HTTP probe (For Hono/Next.js)
   curl -sf https://{domain}/health || curl -sf https://{domain}/api/health
   Successful if it returns 200.

5. If failed → analyze logs and find the root cause

Bun/Hono health check endpoint assumption:

GET /health → { status: "ok", uptime: X }
GET /api/health → 200 OK

If this endpoint does not exist, ask the user to add it.

Deploy monitoring loop (bash):

for i in $(seq 1 18); do
  STATUS=$(curl -s -H "Authorization: Bearer $COOLIFY_ACCESS_TOKEN" \
    "$COOLIFY_BASE_URL/api/v1/deployments/$DEPLOY_UUID" | jq -r '.status')
  echo "[$i] $STATUS"
  [ "$STATUS" = "finished" ] && break
  [ "$STATUS" = "failed" ] && echo "ERROR: Deploy failed" && break
  sleep 10
done

---

[PROVISIONING] New Service Provisioning

> Detail: references/provisioning.md

Objective: Set up all services from scratch for a tenant/project.

Standard order — strictly follow this order:

1. Create project (or find existing project)
2. Create environment (staging / production)
3. Create database and obtain connection credentials
4. Create application (source: Docker image / GitHub repo)
5. Set env vars (including DB_URL)
6. Deploy
7. Run [DEPLOY-VERIFY] loop

Select stack template (ask user if not specified):

| Stack | Template | |-------|----------| | Bun + Hono + PostgreSQL | references/provisioning.md#bun-hono-pg | | Next.js + PostgreSQL + Redis | references/provisioning.md#nextjs-pg-redis | | PostgreSQL only | references/provisioning.md#db-only |

Pre-provisioning data collection from user:

• Project/tenant name
• Target environment (staging/production)
• Source (Docker image tag or GitHub repo + branch)
• Domain (for Traefik)

---

[MULTI-TENANT-ENV] Multi-Tenant Env Var Management

> Detail: references/multi-tenant-env.md

Objective: Manage env vars securely and isolatedly for multiple apps/tenants.

Basic rules:

1. Each tenant's env vars should go only to their app — strictly no cross-contamination
2. You must always restart after changes
3. Do not show secrets in the log — apply masking

Update env var for single application:

# List
curl -s -H "Authorization: Bearer $COOLIFY_ACCESS_TOKEN" \
  "$COOLIFY_BASE_URL/api/v1/applications/$APP_UUID/envs"

# Add/update
curl -s -X POST -H "Authorization: Bearer $COOLIFY_ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"key":"DATABASE_URL","value":"...","is_multiline":false}' \
  "$COOLIFY_BASE_URL/api/v1/applications/$APP_UUID/envs"

In MCP mode: env_vars(resource: application, action: create|update, uuid, key, value)

Bulk update (multiple applications): If MCP is present: bulk_env_update(app_uuids: [...], key, value) — write to all apps with a single command. In API mode: run the references/multi-tenant-env.md#bulk-script script.

Verify: isolation check List the env vars of each app, ensure that DATABASE_URL values are different.

---

[CONTAINER-EXEC] Container Exec / SSH Fallback

> Detail: references/container-exec.md

When to use:

• When API is insufficient (file editing, config patch, log file reading)
• When .env / config file needs to be edited inside the container
• In-container debug (node_modules, bun cache, etc.)

Access strategy (try in order):

1. Coolify Dashboard → Service → Terminal (web UI) — easiest
2. SSH → server → docker exec — if shell access is available
3. Coolify API exec endpoint (if available) — programmatic

SSH + docker exec:

# Connect to server
ssh user@coolify-server

# Find container name
docker ps | grep 

# Enter container
docker exec -it  sh
# or if bash exists:
docker exec -it  bash

File path assumptions in Coolify:

| Service | File path | |--------|------------| | Bun/Node app | /app/ or /usr/src/app/ | | Next.js | /app/ | | PostgreSQL data | /var/lib/postgresql/data/ |

After in-container operations are completed, always restart the service:

# API mode
curl -X POST -H "Authorization: Bearer $COOLIFY_ACCESS_TOKEN" \
  "$COOLIFY_BASE_URL/api/v1/applications/$APP_UUID/restart"

---

General Principles

Token efficiency: First use summary endpoints, if detail is needed go to single source.

Order in case of error:

1. Fetch logs (last 200 lines is enough)
2. Check status
3. Look at references/ files for known error patterns
4. Ask user one clarifying question — not multiple

Security:

• Never show API token in log or output
• Mask env var values: DATABASE_URL=postgres://***
• Obtain user approval prior to destructive operations (stop_all, delete)

Coolify API base pattern:

curl -s \
  -H "Authorization: Bearer $COOLIFY_ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  "$COOLIFY_BASE_URL/api/v1/{endpoint}"

---

Reference Files

Read the relevant file in the following cases:

| File | When | |-------|----------| | references/turborepo-deploy.md | Turborepo/monorepo build issues | | references/deploy-verify.md | Deploy loop details, Bun/Hono/Next.js health check patterns | | references/provisioning.md | New service provisioning templates | | references/multi-tenant-env.md | Bulk env update, isolation verification | | references/container-exec.md | SSH patterns, file editing, container debug | | references/api.md | All Coolify REST API endpoints (if MCP unavailable) |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: fatih-developer
• Source: fatih-developer/fth-skills
• License: MIT

Install and usage instructions live in the source repository linked above.