generated: '2026-08-12'
method: searched
source: >-
  https://developers.skai.io/ — "Rate Limits" section of the Skai OpenAPI info.description
  (openapi/skai-kenshoo-api-openapi.yml)
limit_count: 2
scope_note: >-
  Limits are per API USER, not per key, per account or per endpoint. Because the refresh
  token is tied to a user, several integrations sharing one service user share one budget —
  Skai's own docs recommend creating a dedicated user for API requests partly for this reason.
limits:
- scope: per-user
  window: 1 minute
  limit: 60
  burst: null
  applies_to: all Skai REST API endpoints
- scope: per-user
  window: 1 hour
  limit: 2000
  burst: null
  applies_to: all Skai REST API endpoints
exhaustion:
  status_code: 429
  message: API rate limit exceeded
  retry_after_header: null
  retry_after_note: >-
    Skai does not document a Retry-After header on 429, and 429 is not declared as a response
    on any of the 158 operations in the published OpenAPI — an agent reading the contract
    alone would never anticipate it.
response_headers:
  documented: true
  named: false
  statement: >-
    "When calling any API endpoint the response headers will show the limits relevant to this
    user, and the number of remaining calls you can make within the current minute/hour."
  header_names: null
  header_names_note: >-
    NOT RECORDED — Skai states that limit and remaining headers are returned but never names
    them. They are neither the RFC 9239 RateLimit-* family nor the X-RateLimit-* convention on
    any published page, and the header block is absent from components.headers in the OpenAPI
    (components.headers is empty). Confirming the actual names requires an authenticated call.
    Nothing is guessed here.
  observed: false
  observed_note: >-
    Could not be observed anonymously: every Skai REST endpoint requires a bearer token, so an
    unauthenticated request returns 401 before any rate-limit header is emitted.
token_endpoint_note: >-
  Access-token minting is itself rate-limited. Skai advises caching an access token for its
  full `expires_in` (typically 21600s / 6h) "to prevent rate limit issues with generating new
  tokens too often."
mcp_limits:
  documented: false
  note: >-
    No rate limits are published for the hosted MCP servers at mcp.kenshoo.com. Skai's blog
    claims MCP reporting takes "roughly one-tenth the number of calls" of a publisher-native
    interface, but publishes no MCP quota.
mitigations_documented:
- Use asynchronous analysis reports for datasets over a few thousand rows instead of polling
  the synchronous endpoint.
- Filter reports to non-zero metrics (e.g. impressions > 0) to shrink result sets.
- Scope structure reports with a "last updated" filter to fetch only recently changed entities.
- Use Bulk Update (one file) for large-scale entity changes rather than per-entity calls.