generated: '2026-08-13'
method: searched
source: https://www.flint.com/docs/api
docs: https://www.flint.com/docs/api
authentication:
  rest: "Bearer API key (Authorization: Bearer ak_...) — see authentication/flint-authentication.yml"
  mcp: "OAuth 2.0 via Clerk — see authentication/flint-authentication.yml"
  key_provisioning: >-
    Created in Flint team settings; scoped to the organization; requires at
    least `member` role permissions.
async_execution:
  model: >-
    Agent tasks are asynchronous. createAgentTask returns a taskId with
    status "running"; the caller either polls getAgentTask or supplies a
    callbackUrl to be notified on completion.
  states: [running, completed, failed]
  identifier: taskId
  identifier_format: "bg-<site-uuid>-<suffix>"
  mcp_identifier_alias: >-
    The MCP tools call the same identifier `workflowId` (run_background_agent
    returns a "workflow ID"; check_background_agent_status takes `workflowId`).
    Same value, different name across the two surfaces.
  polling_guidance:
    interval: 90-120 seconds between status checks
    typical_duration: 4-5 minutes for a single agent
    concurrency_note: ~10 concurrent agents may take ~15 minutes
    source: https://github.com/tryflint/claude-code-plugin/blob/main/skills/edit-site/SKILL.md
publish_semantics:
  field: publish (boolean, default false)
  false: >-
    No production deployment. Changes land on the site's staging branch and
    preview URLs remain accessible.
  true: >-
    Generated pages become immediately visible on the live site and a
    production deployment is triggered.
  standalone_publish: MCP publish_site tool only — no REST equivalent.
  detail: See sandbox/flint-sandbox.yml
idempotency:
  supported: false
  note: >-
    Flint's Agent Tasks API does not document an idempotency-key header or
    parameter. Each POST /agent/tasks starts a new background task, and each
    task consumes credits — so a blind retry on a network timeout both
    duplicates work and double-charges. Poll before retrying.
pagination:
  supported: false
  note: No list endpoints with pagination are documented in the REST API.
field_expansion:
  supported: false
metadata:
  supported: false
request_tracing:
  request_id_header: not documented
payload_limits:
  - operation: createAgentTask (generate_pages)
    field: items
    limit: 1-10 items per task
  - operation: createAgentTask
    field: callbackUrl
    constraint: must be HTTPS
webhooks:
  supported: true
  mechanism: callbackUrl
  transport: HTTPS POST
  trigger: task completion
  subscription: per-request, not per-account
  signing: not documented
  retries: not documented
  payload_fields: [taskId, status, pages, error, timestamp]
  detail: See asyncapi/flint-webhooks.yml
rate_limiting:
  supported: true
  signal: HTTP 429 Too Many Requests
  headers: not documented
  limits_published: false
  note: >-
    Flint states the endpoints are rate limited and that exceeding the limit
    returns 429, but publishes no numbers, windows, scopes or response headers.
    See rate-limits/flint-rate-limits.yml. The binding quota in practice is
    economic (credits), not temporal.
versioning:
  scheme: uri-path
  current: v1
  base_url: https://app.tryflint.com/api/v1
  changelog: none published for the API — see changelog/flint-changelog.yml
error_handling:
  rest: >-
    Plain HTTP status codes with a bare {"error": "<message>"} JSON body — not
    RFC 9457 application/problem+json. See errors/flint-problem-types.yml.
  async: Failed tasks return an errorMessage string on the task-status response.
cross_links:
  errors: errors/flint-problem-types.yml
  lifecycle: lifecycle/flint-lifecycle.yml
  authentication: authentication/flint-authentication.yml
  webhooks: asyncapi/flint-webhooks.yml
  rate_limits: rate-limits/flint-rate-limits.yml
  plans: plans/flint-plans-pricing.yml
  sandbox: sandbox/flint-sandbox.yml