generated: '2026-08-14'
method: searched
source: >-
  openapi/_original/parallel-openapi-original.json + https://docs.parallel.ai
  (rate-limits, warnings-and-errors, webhook-setup, MCP) + live /.well-known
  probes 2026-08-14
authentication:
  style: api-key
  header: x-api-key
  env_var: PARALLEL_API_KEY
  oauth:
    surfaces: [Task MCP, A2A agent, CLI login]
    issuer: https://platform.parallel.ai
    scopes: [key:read]
    pkce: S256
    dynamic_client_registration: true
  docs: https://docs.parallel.ai/getting-started/overview
  see: authentication/parallel-authentication.yml
idempotency:
  supported: false
  notes: >-
    No Idempotency-Key header or parameter is documented or present in the
    OpenAPI. Re-POSTing a Task run creates a new billable run. The closest safety
    property Parallel offers is billing-side, not request-side: failed Task runs
    and unsuccessful Responses calls are not charged.
pagination:
  style: cursor
  notes: List/event endpoints (monitors, monitor events, task-group runs, FindAll results) use cursor-based paging.
streaming:
  style: server-sent-events
  content_type: text/event-stream
  surfaces: [task run events, task group events, FindAll events, monitor events]
  docs: https://docs.parallel.ai/task-api/task-sse
webhooks:
  standard: Standard Webhooks
  signature: HMAC-SHA256 over "<webhook-id>.<webhook-timestamp>.<payload>"
  headers: [webhook-id, webhook-timestamp, webhook-signature]
  secret_prefix: whsec_
  docs: https://docs.parallel.ai/resources/webhook-setup
  see: asyncapi/parallel-webhooks.yml
versioning:
  scheme: uri-path
  current: v1
  stages: [v1 (GA), v1beta (beta), v1alpha (legacy)]
  see: lifecycle/parallel-lifecycle.yml
error_envelope:
  format: application/json
  shape: '{"error":{"message":"...","detail":{}}}'
  problem_json: false
  retryability_published: true
  see: errors/parallel-problem-types.yml
rate_limiting:
  scope: per-endpoint, POST-create only (GETs are not counted)
  response_headers: none published
  exhaustion_status: 429
  docs: https://docs.parallel.ai/getting-started/rate-limits
  see: rate-limits/parallel-rate-limits.yml
request_tracing:
  request_id_header: null
  notes: >-
    No request-id/correlation header is documented. On the MCP surface an
    agent-supplied `session_id` argument correlates tool calls in Parallel's logs
    and drives free-tier rate limiting, which is the nearest equivalent.
agent_conventions:
  markdown_docs: >-
    Every docs page is retrievable as markdown by appending `.md` to the URL or
    sending `Accept: text/markdown`; the index is at
    https://docs.parallel.ai/llms.txt and the full text at
    https://docs.parallel.ai/llms-full.txt.
  agent_card: a2a/parallel-agent-card.json
  agent_skills: skills/_index.yml
  mcp: mcp/parallel-mcp.yml
  crosswalk: mcp/parallel-tool-crosswalk.yml
  payments:
    protocols: [Machine Payments Protocol, x402]
    docs: https://docs.parallel.ai/integrations/agentic-payments
output_conventions:
  basis: >-
    Every Task Run result carries a `basis` array with citations, reasoning and
    per-field confidence. Requesting those as output-schema fields raises a
    warning — read output.basis instead.
  docs: https://docs.parallel.ai/task-api/guides/access-research-basis