Spree API V3

Spree API v3

Spree exposes two distinct API surfaces under /api/v3/:

They share conventions (envelope shape, prefixed IDs, pagination) but have different auth, different default actions, and different exposed fields. This is the most-confused-about distinction in the API; cover it carefully.

Store API vs Admin API — the contract differences

Store API

Who calls it: customer browsers and apps. Public, untrusted clients.

Auth: Always include X-Spree-Api-Key: pk_<token> (a publishable key). Additional layers:

• Anonymous browse: publishable key alone is enough for reading products, categories.
• Guest cart: publishable key + X-Spree-Token: <cart_token> for operations on a specific guest cart.
• Logged-in customer: publishable key + Authorization: Bearer <jwt> for account data, order history.

What's exposed: customer-visible fields only. Catalog and order Store serializers omit created_at/updated_at, exposing business dates instead (available_on, completed_at); a few resources (digitals, newsletter subscriptions) do include timestamps. No cost prices, no admin internal notes, no private metadata.

What actions are enabled: read-only by default. index and show for catalog endpoints. Cart/customer/address endpoints opt into create/update/destroy.

Channel scope: Store responses are scoped by X-Spree-Channel: <code> (e.g. online, pos). When omitted, the store's default channel is used. Products not published on the requested channel don't appear.

curl -H "X-Spree-Api-Key: pk_CzEKBTWFiuNLgz4wciLsS59n" \
     -H "X-Spree-Channel: online" \
     -H "Accept-Language: en-US" \
     https://my-spree.example.com/api/v3/store/products

Admin API

Who calls it: trusted backend apps (your ERP integration, marketplace fulfillment service) or trusted humans (admin SPA users). Never the browser of an anonymous visitor.

Auth: Two distinct paths, each with its own authorization model.

Path 1: Secret key with scopes — for server-to-server apps and integrations.

curl -H "X-Spree-Api-Key: sk_…" \
     https://my-spree.example.com/api/v3/admin/orders

Secret keys (sk_* prefix) carry scopes that gate which endpoints they can hit. Scopes are granted at key creation. Each request's scope is enforced by ScopedAuthorization; missing scope = 403.

The scope list (5.5, from Spree::ApiKey::SCOPES):

read_orders               write_orders
read_products             write_products
read_promotions           write_promotions
read_customers            write_customers
read_payments             write_payments
read_fulfillments         write_fulfillments
read_refunds              write_refunds
read_gift_cards           write_gift_cards
read_store_credits        write_store_credits
read_stock                write_stock
read_categories           write_categories
read_settings             write_settings
read_webhooks             write_webhooks
read_api_keys             write_api_keys
read_dashboard
read_all                  write_all     # superset (full admin)

Some endpoints map onto these rather than having their own pair: custom-field-definition endpoints require read_settings/write_settings, and export endpoints resolve their required scope per request (there is no read_exports/write_exports scope).

This is the right path for building an app or integration. The app gets a minimum-privilege secret key from the merchant; no human user is involved. Audit-friendly (you know exactly which app made each request).

Path 2: JWT with CanCanCan abilities — for human admin users.

# Login first
curl -X POST https://my-spree.example.com/api/v3/admin/auth/login \
     -H "Content-Type: application/json" \
     -d '{"email":"admin@example.com","password":"…"}'
# Returns: { token, user } — the JWT is in `token`; the refresh token is set as an
# httpOnly cookie (used by POST /api/v3/admin/auth/refresh), not returned in the body.

# Then use the JWT
curl -H "Authorization: Bearer <jwt>" \
     https://my-spree.example.com/api/v3/admin/orders

JWT admin auth uses Spree::Ability (CanCanCan) to determine what the human user can do. Roles + permission sets configure who can manage what.

(The store is resolved from the request host, not from an API key.)

What's exposed: everything visible to the Store API plus timestamps (created_at, updated_at, deleted_at if paranoid), cost prices, private metadata, internal notes, audit fields (approved_by_id, cancelled_by_id), back-office relations.

What actions are enabled: full CRUD by default. index, show, create, update, destroy for every resource unless explicitly restricted.

Quick reference

The {data, meta} envelope

Every list endpoint returns:

{
  "data": [
    { "id": "prod_86Rf07xd4z", "name": "...", ... },
    { "id": "prod_kvJ0pQrTb9", "name": "...", ... }
  ],
  "meta": {
    "page": 1,
    "limit": 25,
    "count": 152,
    "pages": 7,
    "from": 1,
    "to": 25,
    "in": 25,
    "previous": null,
    "next": 2
  }
}

Single-record endpoints return the record's attributes directly (no wrapping):

{ "id": "prod_86Rf07xd4z", "name": "...", ... }

Conventions:

• There is no type field. The resource kind is implied by the prefixed-ID prefix (prod_, or_, variant_, …). A few resources do expose a type attribute (payment methods, promotion rules/actions, price rules), but it is an STI class discriminator specific to that resource, not an envelope convention.
• meta is on lists only. Single-record responses don't have it.
• next / previous are page numbers (or null at the ends). Pagination is offset-based via Pagy — pass ?page=N&limit=N to navigate.

Prefixed IDs

Every v3 API uses Stripe-style prefixed IDs:

prod_86Rf07xd4z       Product
variant_k5nR8xLq      Variant
or_m3Rp9wXz           Order
py_…                  Payment       (Stripe parity)
ful_…                 Shipment
adj_…                 Adjustment
li_…                  LineItem
ch_…                  Channel
key_…                 ApiKey        (record ID; the credential token values are prefixed pk_/sk_ — those are secrets, not IDs)
cf_…                  CustomField   (alias of Metafield)

The integer PK is never in API responses — only the prefixed form. Same on writes: send "variant_id": "variant_k5nR8xLq", not "variant_id": 42. The server resolves prefixed IDs to integer PKs internally.

Computation: Sqids.encode([integer_pk]) with min_length: 10. Deterministic — the same PK always gets the same prefixed ID. There's no database column for it; it's computed on read.

Expand

Resources support an expand query param for sideloading related data:

curl '/api/v3/store/products/cool-shirt?expand=media,default_variant,categories'

Returns the product with media, default_variant, and categories inlined as full objects. Without expand, related objects appear as ID references on the parent. Dot notation lets you expand nested associations:

curl '/api/v3/store/products/cool-shirt?expand=variants.media'

Allowed expand keys are per-resource and listed in the OpenAPI spec.

Sparse fieldsets

The inverse of expand: pass fields=name,slug to trim the response to just those attributes (id and any expanded associations are always retained):

curl '/api/v3/store/products?fields=name,slug'

Pagination

GET /api/v3/admin/orders?page=2&limit=50

Offset-based via Pagy. limit defaults to 25; max is generally 100. Use meta.next / meta.previous to navigate.

Filtering with Ransack

List endpoints accept Ransack predicates as q[<attribute>_<predicate>]:

# Products with name containing "shirt"
GET /api/v3/store/products?q[name_cont]=shirt

# Orders completed in the last 30 days
GET /api/v3/admin/orders?q[completed_at_gteq]=2026-05-01

# Multiple filters — use `status` (draft/placed/canceled, new in 5.5), not the legacy `state` (removed in Spree 6)
GET /api/v3/admin/orders?q[status_eq]=placed&q[total_gt]=100

# Sort by completed_at descending
GET /api/v3/admin/orders?q[s]=completed_at+desc

Common predicates: _eq, _not_eq, _in, _not_in, _cont (LIKE %x%), _start (LIKE x%), _gteq, _lteq, _gt, _lt, _present, _blank.

Only attributes in the model's whitelisted_ransackable_attributes and whitelisted_ransackable_associations are queryable. Predicates on attributes outside the whitelist are silently ignored — the request returns 200 with that condition dropped (any remaining whitelisted predicates still apply), not a 422.

Error responses

All errors use a consistent envelope:

// 401 Unauthorized
{ "error": { "code": "invalid_token", "message": "Valid API key required" } }

// 403 Forbidden (scope missing)
{ "error": { "code": "access_denied", "message": "API key lacks scope: write_orders", "details": { "required_scope": "write_orders" } } }

// 404 Not Found
{ "error": { "code": "record_not_found", "message": "Product not found" } }

// 422 Unprocessable Entity (validation)
{
  "error": {
    "code": "validation_error",
    "message": "Email can't be blank and Customer is required for checkout",
    "details": {
      "email": ["can't be blank"],
      "base": ["Customer is required for checkout"]
    }
  }
}

// 429 Too Many Requests (rate limited)
{ "error": { "code": "rate_limit_exceeded", "message": "..." } }

Note: some resources return a specific code instead of record_not_found: order_not_found (orders), cart_not_found (carts — order lookups on /carts paths), line_item_not_found, variant_not_found. The message is always " not found".

The details map on 422s uses attribute names as keys. Map field-level errors to inputs; base errors are non-field-specific (display as a form-level banner). The @spree/sdk includes a SpreeError class that parses these automatically.

Rate limiting

The API has per-key and per-endpoint rate limits configured via Spree::Api::Config:

Hit limits → 429 Too Many Requests with Retry-After header. The @spree/sdk retries with exponential backoff automatically.

Tune via Spree::Api::Config[:rate_limit_per_key] etc. in config/initializers/spree.rb. For tougher global throttling (per-IP at the proxy edge), layer Rack::Attack or your CDN's WAF on top.

Read/write attribute symmetry (a v3 invariant)

For any resource: whatever a serializer returns, the controller's permitted_params accepts on write under the same name. No label exposed but presentation accepted. No customer_note exposed but special_instructions accepted. The client never has to translate.

When the underlying column has a legacy name, the model has an alias method (e.g. Spree::OptionType exposes label/label= aliasing the underlying presentation column). The model owns the bridge.

Common debugging recipes

"I'm getting 401 on every call"

• Check X-Spree-Api-Key header is set.
• Verify the key is valid: GET /api/v3/admin/me returns 200 if your JWT is valid; otherwise 401.
• Publishable key for Store API; secret OR JWT for Admin API. Mixing them = 401.

"I'm getting 403 on Admin API"

• For secret keys: the key doesn't have the required scope. Check the key's scopes; either grant the scope or use a JWT admin user.
• For JWT admins: the user doesn't have the required ability. Check Spree::Ability rules for the user's role.

"My q[...] filter is silently ignored"

The attribute isn't in the model's Ransack allowlist. The API uses lenient .ransack, so conditions on non-whitelisted attributes are silently dropped — the list comes back unfiltered (200, no error). Add the attribute via Spree.ransack.add_attribute(Spree::Product, :attr) in an initializer, or append it to whitelisted_ransackable_attributes in a model decorator.

"Empty data array but I know records exist"

• Wrong channel? Store API queries are channel-scoped. Try X-Spree-Channel: <code> matching where the record was created.
• Wrong currency? Some resources (products) are filtered by available(currency) — record exists but doesn't have a price in the requested currency.
• Authorization scope? For Admin: maybe your role can see the index but the scope of accessible records is restricted.

"Webhook payload uses raw IDs?"

It doesn't — Webhooks 2.0 uses the same prefixed IDs as the API. If you're seeing integer IDs, you're either on the legacy webhooks system OR the consumer is parsing wrong.

Where to read further

• OpenAPI specs: node_modules/@spree/docs/dist/api-reference/store.yaml (Store API) and admin.yaml (Admin API) — every endpoint, parameter, response schema. Authoritative.
• Adding a new endpoint: see the spree-resource skill — spree:api_resource generator produces v3-conformant controllers + serializers automatically.
• SDK: @spree/sdk (Store) and @spree/admin-sdk (Admin) — typed clients. See the spree-typescript-sdk skill.
• Customization: node_modules/@spree/docs/dist/developer/customization/api.md and authentication.md
• Webhooks vs subscribers: see the spree-events-webhooks skill.