generated: '2026-09-20'
method: searched
source: https://maxvideoai.com/docs/mcp
surface: MCP (Model Context Protocol, streamable-http, JSON-RPC 2.0)
authentication:
  style: OAuth 2.1 bearer (header), PKCE S256, dynamic client registration
  cross_link: authentication/maxvideoai-authentication.yml
idempotency:
  coverage: partial
  scope:
  - confirm_generation
  mechanism: >-
    The paid write is a two-phase quote/approve flow: prepare_generation returns
    an exact short-lived quote (with expiresAt) and confirm_generation submits
    that quote idempotently. Re-issuing the same confirmed quote does not create a
    second paid attempt; one confirmation authorizes exactly one paid attempt and
    is consumed whether the job is accepted, failed, or refunded.
  header: null
  note: >-
    Idempotency is keyed on the server-issued quote id, not a client-supplied
    Idempotency-Key header, and covers only confirm_generation. Read tools and
    the import/upload/topup handoffs are not described as idempotent.
reversibility:
  state: documented
  detail: >-
    There is no agent-invokable cancel/refund/undo tool. Generation is one-way:
    a completed paid generation cannot be reversed. Failed generations are refunded
    automatically and get_generation_status returns the refund/recredit state; a
    refund does not restore the paid authorization (a retry needs a fresh quote and
    approval). No reversal window is stated for a successful generation.
  reversal_operations: []
  window: null
pagination:
  style: null
  note: Not documented for the MCP tool surface; result-set tools (list_media, list_recent_generations) return bounded lists.
versioning:
  scheme: semver
  current: 0.3.5
  cross_link: changelog/maxvideoai-changelog.yml
request_tracing:
  correlation_id: correlationId
  note: INTERNAL_ERROR responses return a correlationId to retain when contacting support.
error_envelope:
  style: JSON-RPC 2.0 error object + named error codes
  cross_link: errors/maxvideoai-problem-types.yml
rate_limit_signaling:
  note: >-
    Docs state "server limits apply" but publish no rate-limit headers or numeric
    limits; authenticated responses carry Cache-Control: private, no-store.
  cross_link: rate-limits/maxvideoai-rate-limits.yml
dry_run_mode:
  state: na
  note: >-
    prepare_generation / calculate_project_budget are non-charging planning and
    quoting steps that rehearse a request without spending, but there is no
    per-call dry-run flag on the paid confirm.