Lamina Labs Agentic Access

generated: '2026-07-19'
method: generated
source: mcp/lamina-labs-mcp.yml
model: x-agentic-access (Curity Access Intelligence vocabulary)
disclaimer: A governance starting point derived from the provider's published MCP tool
  annotations and OAuth scopes, not an authoritative provider claim. `audience` is left
  null to bind per deployment.
summary:
  operations: 2
  by_action_class: {connected: 1, acting: 1}
  by_consequence: {read: 1, write: 1}
  human_in_the_loop: 0
  audit_required: 1
operations:
- operation: simi_get_video
  surface: mcp
  scope: jobs:read
  x-agentic-access:
    action-class: connected
    consequence: read
    scope: jobs:read
    audience: null
    token: {max-ttl: 3600}
    audit: optional
    human-in-the-loop: not-required
  rationale: Provider annotates this tool readOnlyHint true, idempotentHint true,
    destructiveHint false. It returns signed media URLs, so the response is sensitive
    even though the call is read-only.
- operation: simi_submit_video
  surface: mcp
  scope: jobs:create
  x-agentic-access:
    action-class: acting
    consequence: write
    scope: jobs:create
    audience: null
    token: {max-ttl: 900}
    audit: required
    human-in-the-loop: not-required
  rationale: Creates a billable video generation job. Provider annotates idempotentHint
    false, so a retry produces a second job; a governing agent should de-duplicate and
    treat repeat submission as a cost event.
  cautions:
  - No idempotency key exists, so at-least-once agent retry semantics will double-charge
    and double-generate.
  - openWorldHint is true on both tools; outputs are model-generated video and should not
    be treated as authoritative content.
notes:
- The provider's authorization server supports dynamic client registration with the
  `none` token endpoint auth method, so an agent can self-register as a public client.
  Deployments should constrain which agents may complete that flow.