Skip to main content
REST and MCP use the same account credits. Limits apply at different levels: API-key requests, brand onboarding, and unfinished AI Gateway generations.

​
Rate limits

​
API-key requests

Each API key has a default limit of 120 accepted requests. The count resets 60 seconds after the last accepted request. Requests rejected by this limit do not delay the reset. REST and MCP requests using the same key count toward the same limit, whether you send the key in x-api-key or Authorization: Bearer. OAuth tokens are not subject to this API-key limit. When you reach the limit, requests return HTTP 429. Wait for the number of seconds in the Retry-After response header before retrying. If the header is missing, wait 60 seconds. REST uses code TOO_MANY_REQUESTS; MCP uses rate_limit_exceeded. REST responses to API-key requests include a policy header. For the default quota:
q is the key’s request limit. w is how many seconds must pass after the last accepted request before the count resets. This header does not report remaining requests or a reset timestamp. Invalid keys and unmatched routes can receive the default policy; requests without credentials, OAuth requests, and the public specification omit it. For a higher request quota, contact support.

​
Brand onboarding

Each personal account or team can start 5 new brand onboardings in a rolling 60-second window, shared across members, API keys, and OAuth tokens using REST or MCP. Replaying an accepted idempotent request and reading progress do not consume this onboarding budget. The API-key request limit applies separately. At this limit, REST returns TOO_MANY_REQUESTS with Retry-After; MCP returns RATE_LIMITED with retry_after_seconds. Wait for the returned delay.

​
AI Gateway generation

An account without an active paid plan or positive purchased credits can have 3 unfinished billed generations at once. A paid plan granted at no charge also removes this limit. A further REST request returns ACTIVE_GENERATION_LIMIT_REACHED with the oldest unfinished generation’s ID and a pricing link. MCP returns the same guidance in a tool error. Wait for a generation to finish or fail before starting another, or upgrade. An unfinished generation stops occupying its slot after two hours.

​
Credits

Use Get credit balance to read the balance and credit_costs for the personal account or team you are using. Those returned costs account for legacy pricing. List workspaces identifies the teams you can access. Current plans, credit allowances, and top-up options are on Pricing. Plan credits renew monthly on the billing anniversary and do not roll over. Purchased top-up credits do not expire.

​
AI Gateway billing

Any positive, usable credits with a known value can fund a billed generation, including signup credits. Before starting, Rainbrand reserves the estimated charge plus a 10% buffer. The account needs enough available credits for that reservation and no outstanding debt. After completion, Rainbrand settles the actual charge and returns unused reserved credits. If the charge exceeds the reservation, remaining credits cover the difference; an unpaid amount becomes debt. Failed or unusable output is not charged. Reservations left by interrupted jobs expire within four hours. Completed output remains available while billing retries in the background. Models that are not billed can run without a charge.

​
Billing errors

Use error.code to decide what to do. The HTTP status alone is not enough. MCP returns the corresponding guidance in tool errors. If a credit balance read says the billing account is not set up, contact support with the billing ID in the message.