generated: '2026-09-03'
method: searched
source: https://imaginode.ai/en/docs/api-mcp
note: >-
  Imaginode publishes no OpenAPI; these cross-cutting semantics are read from the API & MCP
  documentation and from live probes of the three documented REST endpoints. The write
  surface is exactly one operation (POST /api/generate) plus its MCP tool equivalents.
authentication:
  style: bearer API key (imk_ prefix) in the Authorization header
  artifact: authentication/imaginode-authentication.yml
idempotency:
  coverage: none
  note: >-
    No idempotency mechanism (Idempotency-Key header or equivalent) is documented for
    POST /api/generate or the MCP generation tools. A retried request starts and bills a
    new generation. The 10-generations-per-minute per-key limit is the only documented
    guard against a runaway or retrying agent.
pagination:
  style: none
  note: GET /api/models returns the full model catalog in one response; no paginated endpoints are documented.
async_pattern:
  style: polling
  note: >-
    POST /api/generate returns requestId, cost and the new credit balance immediately;
    poll GET /api/generate/status?requestId=... (states: running, completed with output
    URLs, failed with refund). Docs recommend polling every 30 seconds for video; expect
    1-5 minutes typically, up to 30 for long generations. MCP generate_image waits inline
    and returns the image URL directly; generate_video returns a request_id for polling.
cost_transparency:
  note: >-
    The credit cost of a generation is returned at launch, before the result. GET /api/models
    (unauthenticated) carries the per-model credit price list.
error_envelope:
  shape: '{"error": "<message>"}'
  note: 'Observed on live 401 responses: {"error": "Non authentifie"} (messages may be localized in French). See errors/imaginode-problem-types.yml.'
rate_limit_signaling:
  documented_limit: 10 generations per minute per API key
  headers: none documented
  artifact: rate-limits/imaginode-rate-limits.yml
versioning:
  style: none documented
  note: No API versioning, deprecation policy or changelog is published for the API surface.
request_tracing:
  note: No request-id response header is documented; requestId identifies generation jobs, not HTTP requests.
reversibility:
  grade: none
  write_surfaces:
    - operation: POST /api/generate (and MCP generate_image / generate_video)
      reversal: none
      window: null
      note: >-
        No cancel, void or undo operation is documented for a started generation; a launched
        job runs to completion or failure. The provider automatically refunds credits when a
        generation fails (provider crash or timeout, refunded within seconds) - this is a
        provider-initiated refund on failure, not a caller-invoked reversal, so no
        reversibility credit is claimed.
      docs: https://imaginode.ai/en/docs/api-mcp
dry_run_mode:
  supported: none documented
  note: >-
    No dry-run or test mode exists, but the exact credit cost is displayed/returned before
    every generation, and GET /api/models exposes the full price list without authentication -
    an agent can compute cost before committing.