Headless e-commerce backend for an AI-built storefront
Short answer: if you built a storefront with an AI tool and it needs products, carts, orders and an admin for staff, run Medusa as the backend. Medusa is a headless commerce platform: a Node.js server exposes Store and Admin APIs, and any frontend can sell through it. The Tracklayer template deploys Medusa 2.21 on Railway with its admin dashboard, a separate worker, Meilisearch, Postgres, Redis and a bucket for images. It also deploys the official Next.js starter storefront, which you can keep, replace or use as a reference.
Deploy Medusa on RailwayWhen this makes sense
An AI-built storefront is mostly screens. A shop also needs a catalogue, prices per region, stock, carts, orders, taxes and an admin that non-developers can use. Rebuilding that logic in generated code is slow and easy to get wrong. A headless backend keeps those rules in one place while the frontend stays yours.
Use cases listed in the template documentation:
- a direct-to-consumer shop with a custom Next.js storefront,
- a backend for a mobile app, marketplace or B2B portal,
- multi-region stores with separate currencies, taxes and fulfilment settings,
- a proof of concept before moving off a hosted e-commerce platform.
If you only need a content backend (pages, articles, media) rather than a shop, see the Directus guide instead.
What the template deploys
Seven services run in one Railway project:
| Service | Purpose | Public |
|---|---|---|
| Backend | Store and Admin API, admin dashboard at /app, runs migrations before each deploy |
Yes |
| Worker | Subscribers, scheduled jobs, workflows and search indexing | No |
| Storefront | Official Next.js 15 starter shop | Yes |
| Media-Proxy | Read-only server for images in the private bucket | Yes |
| Meilisearch | Product search, on a volume | No |
| Postgres | Database | No |
| Redis | Sessions, events, workflows, locking, cache | No |
Search stays private: the storefront calls Medusa’s POST /store/search, so the Meilisearch key never reaches the browser. If you clear MEILISEARCH_HOST on the Backend and Worker, Medusa falls back to its built-in Postgres search.
Deploy it, step by step
- Choose a paid plan. The template notes recommend Hobby or higher. Trial’s 1 GB per service is likely to run out during imports and admin use.
- Set
MEDUSA_ADMIN_EMAILon the Backend to your own address before deploying. No other input is required. - Deploy and wait. The Backend pre-deploy step runs migrations and creates a demo store. The Storefront waits for the Backend on its first start, so the first deploy takes a while.
- Open the admin at
https://<backend-domain>/app. Sign in with your email andMEDUSA_ADMIN_PASSWORDfrom the Backend Variables tab. Change the password. - Open the Storefront domain. The demo region and products should be live. The storefront receives its publishable API key from the Backend over the private network at start, so nothing is copied by hand.
- Replace the demo data with your own products, regions and prices in the admin.
Connecting your own storefront
The included storefront shows how a frontend talks to Medusa. If you keep your AI-built frontend instead:
- Point it at the Backend’s public URL and call the Store API.
- Add its domain to
STORE_CORSandAUTH_CORSon the Backend. Both accept comma-separated origins. Without this, browser calls are blocked. - The template only documents how the key reaches its own storefront. For a separate frontend, follow Medusa’s documentation on publishable API keys.
- Uploaded files get public URLs from the Media-Proxy through
S3_FILE_URL.
After adding custom domains, the template notes list the variables to update: STORE_CORS, ADMIN_CORS, AUTH_CORS and MEDUSA_BACKEND_URL on the Backend, and NEXT_PUBLIC_MEDUSA_BACKEND_URL and NEXT_PUBLIC_BASE_URL on the Storefront.
What it costs
The figures below are estimates from the template’s TEMPLATE.md for a small store (a few hundred products, light traffic). They were not measured on Railway. They use Railway prices of $10 per GB of RAM per month, $20 per vCPU per month, $0.15 per GB of volume per month and $0.015 per GB of bucket storage, with 2 GB assumed in the bucket. Network egress is not included.
| Load | RAM | vCPU | Estimated monthly cost |
|---|---|---|---|
| Idle | about 1.2 GB | about 0.09 | about $14 |
| Typical | about 2.7 GB | about 0.84 | about $45 |
Each extra Worker replica adds about 0.7 GB and 0.2 vCPU, roughly $11 per month at typical load by the same estimate. Your bill depends on traffic, catalogue size and bucket use.
Scaling and upgrades
The Backend, Worker and Storefront are stateless and can run more replicas; raise the Worker’s replica count when background jobs queue up. Meilisearch, Postgres and Redis run as single instances because each owns a volume. The starter is pinned to a commit, so upgrades are deliberate, not automatic.
Limitations and licence notes
- Payments need your own Stripe keys; nothing is charged by default.
- The storefront is the official starter, not a custom design.
- The template is a community template and is not affiliated with or endorsed by MedusaJS, Meilisearch or Vercel.
- Licences are summarised in the FAQ above.
Other options
- Need a database or auth behind a custom app instead of a shop? See Supabase on Railway or the guide to backends for vibe-coded apps.
- For content rather than commerce, use the Directus guide.
Browse all guides.
FAQ
How do I log in to the Medusa admin?
Set MEDUSA_ADMIN_EMAIL on the Backend service to your own address before deploying (the default is admin@example.com). When all services are healthy, open the backend domain at /app and sign in with that email and the generated MEDUSA_ADMIN_PASSWORD from the Backend Variables tab, then change the password.
How do I take payments?
Payments are off by default. Add STRIPE_API_KEY and STRIPE_WEBHOOK_SECRET on the Backend, and NEXT_PUBLIC_STRIPE_KEY on the included storefront, when you are ready. Then enable the Stripe provider per region in the admin.
Is the data kept when I redeploy, and how do I upgrade?
Postgres, Redis and Meilisearch each use a volume, and uploads go to a Railway Bucket. The starter commit and the Meilisearch plugin version are build arguments in the backend and storefront Dockerfiles. Change them together, check the plugin's Medusa compatibility and redeploy; migrations run automatically. Take a dump or snapshot before upgrading Meilisearch.
Are there licence restrictions?
According to the template's NOTICE.md, Medusa and the DTC starter are MIT, and Medusa Enterprise Edition materials are not enabled. Meilisearch Community Edition is MIT; its Enterprise parts (BUSL-1.1) are not used. Redis 8 is offered under RSALv2, SSPLv1 or AGPLv3. If licensing matters for your business, check with a lawyer.