Rate limits
Each API token can send a certain number of requests a minute. Past that, Crumbcast answers
429 and does nothing with the request, so you can send it again once you're told to.
| What's counted | Each minute | At once |
|---|---|---|
| Every request with the token | 600 | 100 |
Its requests that change something (POST, PUT, PATCH, DELETE) | 120 | 30 |
| Its requests that start encoding work: creating or completing an upload, a new format, a clip or trim, captions | 60 | 20 |
A request counts against every row it falls under. Each token has its own limits, so two
tokens of the same project don't share them. A request refused with 429 doesn't count
against any of them.
"At once" is how many you can send in a burst: the limit refills at the per-minute rate, so after a pause you can send a burst again. A script that uploads a large library can start 20 uploads straight away, then about one a second.
What players load isn't counted: playback tokens, manifests, a video's views and its analytics events. Neither is anything your viewers do on your pages.
Every answer says what's left
| Header | Means |
|---|---|
RateLimit-Limit | How many you can send at once, for the limit you're closest to |
RateLimit-Remaining | How many you can send right now |
RateLimit-Reset | Seconds until you can send a full burst again |
RateLimit-Policy | That limit, as per minute;w=60;burst=at once |
Retry-After | On a 429 only: seconds to wait before trying again |
HTTP/1.1 429 Too Many Requests
Retry-After: 2
RateLimit-Limit: 30
RateLimit-Remaining: 0
RateLimit-Reset: 15
RateLimit-Policy: 120;w=60;burst=30
{ "detail": "Too many requests. Try again in 2 s" }
Staying under them
- The SDKs wait for you. The JavaScript, Python
and PHP SDKs send a request again after a
429, waiting as long asRetry-Aftersays (retries, 2 by default). - Use webhooks instead of polling. Asking whether a video is ready every second uses up the limit; a webhook tells you once it is.
- Page through lists rather than fetching each item on its own.
- Agents using the MCP server are held to their token's limits: each tool call counts as the requests it makes.
If your use needs higher limits, ask through the Enterprise contact on Pricing.