Shopify Storefront API throttled / 429 / cost limit killing headless cart — Hydrogen, Oxygen, buyer IP, persisted queries.

Several products can look like “headless cart / PDP died.” They do not share a settings panel. This page is only when Storefront API GraphQL on Hydrogen / Oxygen returns 429 / THROTTLED / cost limit and cart or catalog fetches fail under load. Cart stays empty after cartCreate / cartLinesAdd with cookie / buyer-identity issues is Hydrogen cart broken. Admin GraphQL bulkOperationRunQuery jobs that never finish are bulk operations stuck. Missing Admin / custom app access scopes after an update are custom app scopes missing. This page is only: Shopify Storefront API throttled / 429 / cost limit killing headless cart (Hydrogen, Oxygen, buyer IP, persisted queries).

Six ordinary reasons the headless cart “dies” under traffic: the Storefront query cost is too high (over-fetched fields / nested connections); requests are not using persisted queries; many buyers share one server egress IP so Shopify’s buyer-IP bucket looks empty; you ignore Retry-After and stampede; Oxygen env uses the wrong Storefront token / API version; or you are debugging cart mutation/cookie bugs, Admin bulk ops, or Admin scopes instead of Storefront cost limits. Walk them in that order. You already pay for Shopify + Hydrogen / Oxygen. You do not need another “headless performance SaaS” to stop a 429 you can prove in Network.

Open the live Hydrogen / Oxygen URL (not localhost only). Reproduce Add to cart or a heavy PDP / collection fetch. In DevTools → Network, filter Storefront API GraphQL. Screenshot: HTTP status (429 vs 200), response errors (`THROTTLED`, cost exceeded), extensions.cost (requested / actual / throttleStatus), Retry-After, and whether the request used a persisted query hash vs a raw document.

Note who is calling: browser → Storefront (buyer IP) vs Oxygen server → Storefront (shared egress). Shared server IPs burn the bucket for everyone on that host. If cart mutations succeed with 200 and the cart cookie is present but lines disappear, leave this page and use Hydrogen cart broken. If Admin bulkOperationRunQuery is the failing job, open bulk operations stuck.

Shrink the GraphQL document: drop unused product / metafield / media fields, paginate connections, and avoid N+1 fan-out on every cart line. Prefer persisted queries so Shopify can cache and cost-score known documents. On Oxygen / Hydrogen, confirm the Storefront API token and API version in env match the live shop — wrong token mistakes look like random failures under load.

Respect buyer IP: when the storefront proxies Storefront calls, forward the shopper’s IP per Shopify’s Storefront guidance instead of one shared server IP. On 429, honor Retry-After with exponential backoff — do not retry immediately from every client. If Admin custom app scopes are missing after a reauth while Storefront is fine, that is custom app scopes missing, not this throttle playbook.

Checklist: (1) screenshot live Storefront 429 / THROTTLED / cost extensions, (2) measure requested vs actual query cost on cart + PDP, (3) confirm persisted queries registered and used, (4) confirm buyer IP forwarding vs shared egress, (5) add Retry-After backoff, (6) redeploy Oxygen if token / API version wrong, (7) if mutations/cookies fail with 200, switch to Hydrogen cart broken.

Prove it: under a realistic concurrent load (or production peak), cart and catalog Storefront calls stay under cost / rate limits with 200s, Retry-After is rare and handled, and Add to cart still returns a cart id. If Admin bulk jobs are the only failures, open bulk operations stuck.

Cart empty after successful mutations / lost cookies / buyer identity is Hydrogen cart broken. Admin GraphQL bulkOperationRunQuery jobs that never complete are bulk operations stuck. Missing Admin / custom app access scopes after update / reauth are custom app scopes missing. None of those finish a Storefront API 429 / cost-limit path that kills headless cart under load.

Finish prove 429 → cut cost / persisted queries → buyer IP + backoff proof first. Only then open the related playbook that matches the remaining symptom.

Stay DIY if you can spend one afternoon on the list above: prove Storefront 429 / cost headers, shrink queries and enable persisted queries, fix buyer IP / backoff, and stop treating Hydrogen cart broken / bulk ops / custom app scopes as the same bug. That is the whole playbook. No agency required. Short answers also live on the merchant FAQ. Related pages: Hydrogen cart broken, bulk operations stuck, custom app scopes missing.

Pay for help when the work is not the Storefront cost calculator — it is connecting the stack you already have so lead capture is not dying in DMs, headless / ops exceptions land in a channel you already check, and the live store is one workflow instead of a spreadsheet plus a Slack bot you forgot. That is the Latch AI Ops install.

What $997 USD one-time buys: lead capture on the live store, ops alerts in a channel you already check, and a checkout-ready workflow. 72 hours after collaborator access, not after payment. One live Shopify store. Access plus a delivery thread. Store URL collected at checkout. Not a SaaS seat, not a headless-agency retainer, not a conversion guarantee, not a brand or theme job.

Other shops’ public ranges, not testimonials: a checkout / ops install typically runs $500–$2,500. A full ops consulting block typically runs $2,000–$5,000. $997 is the cheap end of that work, sold as a 72-hour install of three systems — not as an earnings number or uptime claim.

What it does not buy: more orders, automatic Storefront quota, a Hydrogen license, or a case study. Latch AI Ops has not published customer proof because there are no customers yet. You are buying the install.

Checkout is on Whop. Operator is Nico. United States offer. You can revoke collaborator access after handoff. We do not take owner passwords.