generated: '2026-08-09'
method: searched
source: https://rtcstats.com/api-docs.md
docs:
  - https://rtcstats.com/api-docs
  - https://rtcstats.com/api-docs.md
  - https://rtcstats.com/llms.txt
description: >-
  Cross-cutting request/response semantics for the rtcStats REST API v1.0,
  captured from the published API reference and the OpenAPI. The API is small
  and deliberately flat: one bearer token, one version segment in the path, a
  two-field error envelope, and a credit-metered consumption model in place of
  rate limiting.
authentication:
  style: http-bearer
  header: 'Authorization: Bearer <application JWT>'
  scheme_name: BearerAuth
  bearer_format: JWT
  token_issuance: rtcStats dashboard, Settings > Applications; the token is shown once at creation and cannot be read back
  plan_gate: API and MCP access require the Developer plan or above; a 403 is returned when the plan does not include API access
  detail: authentication/rtcstats-authentication.yml
idempotency:
  supported: false
  header: null
  detail: >-
    rtcStats documents no idempotency key. The chunked upload flow does carry a
    client-generated `fileId` shared by every chunk and the assemble request,
    which de-duplicates chunks within one upload, but it is not documented as a
    retry-safe idempotency key and does not protect the credit-consuming
    assemble/analyze step from double submission. Recorded as absent rather
    than reinterpreted.
pagination:
  supported: false
  detail: >-
    GET /v1.0/sessions returns {total, data[]} with no limit/offset/cursor
    parameters. Result-set narrowing is done with the fifteen filter query
    parameters (name, observationTypes, observationTags, os, browser,
    browserVersion, userId, conferenceId, sessionId, hasCritical, hasHigh,
    hasMedium, hasLowScore, hasMediumScore, hasHighScore) rather than by paging.
  filter_combination: Different filters are AND-combined; multiple values within one filter are OR-combined.
  array_parameter_form: repeated query params or comma-separated values
field_expansion:
  supported: false
  detail: >-
    Two fixed projections instead of expansion — POST /v1.0/enrich returns only
    {scores, observationsCount, observations, userAgentData}, while
    GET /v1.0/sessions/{rtcstatsId} returns the full analysis payload.
metadata:
  supported: true
  detail: >-
    userId, conferenceId and sessionId are carried in the rtcstats JWT at
    collection time and become first-class filters on sessions. Session titles
    are set from fileName on upload/analyze.
request_tracing:
  request_id_header: null
  detail: No request-id or correlation header is documented. Responses carry processorVersion (the @rtcstats/rtcstats-processor semver) and, on enrich, generatedAt (epoch ms).
versioning:
  scheme: uri-path
  current: v1.0
  example: https://api.rtcstats.com/v1.0/sessions
  media_type_versioning: false
  header_versioning: false
  payload_schema_version: >-
    The analysis payload carries its own schema version, documented in llms.txt
    (schema 6.3 for the get_session payload; a note that unitless aggregate keys
    were removed at schema 6.9).
  detail: lifecycle/rtcstats-lifecycle.yml
error_envelope:
  format: custom-json
  rfc9457: false
  content_type: application/json
  shape: '{ "error": "<message>", "errorCode": "<code>" }'
  required_fields: [error, errorCode]
  mcp_transport_errors: >-
    The MCP endpoint uses a different shape — a JSON-RPC 2.0 envelope
    (McpStreamableHttpTransportError) for HTTP-level failures, with id always
    null. Tool failures come back with HTTP 200 inside the stream as JSON-RPC
    responses.
  detail: errors/rtcstats-problem-types.yml
rate_limiting:
  request_rate_limits: none documented
  rate_limit_headers: none documented
  model: credit quota
  detail: >-
    Consumption is metered in credits (1 credit = 1 file analyzed), not requests
    per second. Exhaustion is signalled with HTTP 402 "No credits available".
    Remaining balance is readable at GET /v1.0/quota and via the get_quota MCP
    tool. Monthly credits do not roll over; purchased top-up credits do and are
    only drawn on after the monthly allowance is spent.
  see: rate-limits/rtcstats-rate-limits.yml
payload_conventions:
  number_formatting: >-
    Numeric values are rounded — epoch timestamps to a whole millisecond, every
    other number to at most 2 decimals, except audioLevel values which keep more
    precision. Integer counts and large identifiers are preserved.
  omission_vs_null: >-
    The enrich projection omits absent observation fields entirely rather than
    returning null; only `type` and `severity` are always present. In the full
    session payload, aiSummary is null until background generation completes and
    is always null on POST /analyze and on plans without the AI feature.
  plan_gated_fields: embedUrl is omitted from responses on non-Enterprise plans.
  stability_warning: >-
    The docs state that only the documented fields are stable and other fields
    may appear and change without notice.
request_size_limits:
  request_body_cap: ~4.5MB (Vercel platform cap) — larger dumps must use the chunked multipart flow
  assemble_request_cap: at most 1KB
  file_size_limit: plan-dependent; exceeded returns HTTP 413
  chunked_protocol: >-
    multipart/form-data chunks (chunk, fileId, chunkIndex) each returning
    {"success": true} and consuming no credit, followed by a JSON assemble
    request {"assemble": true, "fileId": "...", "fileName": "optional"} which
    runs the pipeline and consumes one credit. Assemble detection is shape-based.
cross_references:
  errors: errors/rtcstats-problem-types.yml
  lifecycle: lifecycle/rtcstats-lifecycle.yml
  authentication: authentication/rtcstats-authentication.yml
  rate_limits: rate-limits/rtcstats-rate-limits.yml
  mcp: mcp/rtcstats-mcp.yml