Skip to content

Rate limits

Limit How much Counted per Over it
Requests 120 a minute key 429 rate_limited, with Retry-After
Failed key attempts 30 a minute sending address 429 rate_limited, Retry-After: 60
New sources 20 a minute workspace 400 bad_request (“Twenty new sources a minute is the limit.”)
MCP calls share the key’s 120 a minute key a JSON-RPC error with HTTP 429

The minute is a fixed window on the clock: it resets at the start of every minute, whenever the key’s first request in it came. Two keys in one workspace each get their own 120.

Every admitted request, including one that then fails with 400, 404, 409 or 422, carries three headers:

Header Means
X-RateLimit-Limit Requests allowed in a window on this key: 120.
X-RateLimit-Remaining Requests left in the current window.
X-RateLimit-Reset When the window resets, in Unix epoch seconds (not seconds from now).
Headers on a real response
{
"request": "GET /api/v1/signals?limit=5",
"status": 200,
"headers": {
"content-type": "application/json",
"cache-control": "no-store",
"x-ratelimit-limit": "120",
"x-ratelimit-remaining": "118",
"x-ratelimit-reset": "1795238160"
}
}

A refused key (401) or a missing scope (403) is answered before the request is counted, so those responses carry no rate headers and cost nothing.

429 Too Many Requests
{
"error": {
"code": "rate_limited",
"message": "120 requests a minute per key. Wait for the window to reset."
}
}

The response carries Retry-After (whole seconds, at least 1) beside the three headers. Wait that long, then carry on. The window then starts again at 120.

  • Page big. limit=200 on /signals reads 200 articles for one request. A month’s backfill takes minutes, not hours.
  • Poll at the speed things change. Articles arrive in cycles of a few minutes; outlooks refresh a few times a day; reports on their schedule. Polling faster returns the same answer.
  • Watch X-RateLimit-Remaining and slow down as it nears zero instead of waiting for a 429.
  • One key per program. A batch job and a live dashboard on separate keys cannot starve each other.

After 30 requests with an invalid key from one address in a minute, further invalid attempts from that address get 429 with Retry-After: 60 instead of 401. It exists to make a misconfigured client back off. It never refuses a valid key, even from the same address, so one broken script in an office does not lock out everyone else.