generated: '2026-07-19'
method: searched
source: https://kive.ai/docs/llms.txt
notes: >-
  Kive ships no public REST API; its cross-cutting semantics are those of the
  Kive MCP server plus the credit/permission model the platform enforces behind
  it. Fields Kive does not publish are recorded as null rather than guessed.
authentication:
  style: oauth2-authorization-code-pkce
  transport: Bearer token in the Authorization header
  artifact: authentication/kive-authentication.yml
  scopes_artifact: scopes/kive-scopes.yml
idempotency:
  supported: false
  header: null
  notes: >-
    Kive documents no idempotency key or replay-safe retry contract. Generation
    is billed per successful call; the documented safety net is a refund rather
    than deduplication ("Credits are only deducted for successful generations.
    Failed generations and content-filtered outputs are automatically
    refunded."). Retrying a generation produces new output and new charges.
pagination:
  style: null
  notes: Not published — MCP tool result shapes are not documented at field level.
versioning:
  scheme: none-published
  notes: >-
    No API version is advertised on the MCP endpoint or in discovery metadata.
    Product changes are announced by date at https://kive.ai/news.
  artifact: changelog/kive-changelog.yml
request_tracing:
  header: null
  notes: Not published.
error_envelope:
  format: mcp-tool-error
  notes: >-
    Per Kive's MCP docs, when a generation cannot run "the client will receive
    the Kive error and can help you choose a different workspace, product,
    preset, or generation setting." No structured error-code registry is
    published.
rate_limiting:
  model: credit-based
  signal_headers: null
  artifact: rate-limits/kive-rate-limits.yml
  notes: >-
    Throughput is governed by workspace credit balance rather than request
    quotas. When monthly credits deplete the user is shown an Upgrade message
    and generation becomes unavailable; free users fall back to 1 free image
    every 20-30 minutes depending on plan.
authorization_model:
  unit: workspace
  notes: >-
    Every MCP action runs under the authorizing user's workspace permissions and
    the workspace credit balance, both enforced server-side. Clients may be
    asked to select a workspace when the user belongs to more than one.
asynchrony:
  model: poll-for-status
  notes: >-
    Generation is long-running; Kive documents that clients "can also check
    generation status while Kive is working" and should request the finished
    asset once ready.
batching:
  video: up to four video presets and up to four sampled outputs in a single request
  images: >-
    Image generation cost is the per-image price multiplied by the number of
    versions requested (default 4).
  bulk: Bulk Generate covers an entire product catalogue in one workflow