generated: '2026-09-05'
method: searched
source: https://agentalog.com/llms-full.txt
docs: https://agentalog.com/api/
cross_links:
  errors: errors/meta-agent-tools-problem-types.yml
  authentication: authentication/meta-agent-tools-authentication.yml
  lifecycle: lifecycle/meta-agent-tools-lifecycle.yml
  rate_limits: rate-limits/meta-agent-tools-rate-limits.yml
authentication:
  style: bearer tokens with kind prefixes (mr_ guest, sess_ session, cred_ prepaid credit) plus x402 X-PAYMENT for agents
  docs: https://agentalog.com/api/
pagination:
  styles:
    - surface: GET /api/listings (public mosaic)
      style: offset
      params: [limit, offset]
      response_fields: [items, limit, offset, next_offset, low_count]
      note: 'No total on purpose: the provider documents that counting tens of thousands per search would be expensive without changing any decision. next_offset is null on the last page.'
    - surface: GET /v0.1/servers (MCP subregistry)
      style: cursor
      params: [limit, cursor]
      response_fields: [servers, metadata.next_cursor, metadata.count]
      note: Official MCP Registry spec v0.1 format — opaque cursor from metadata.next_cursor.
error_envelope:
  shape: '{ ok: false, error: <slug>, message: <pt-BR text>, api_index }'
  format: vendor JSON, not RFC 9457
  observed: 'GET /api/listings/{id} with unknown id -> 404 {"ok":false,"error":"not_found",...} (2026-09-05)'
rate_limit_signaling:
  status: 429
  headers: [Retry-After]
  note: >-
    No X-RateLimit-* headers. Agent-contact backoff (60s doubling to a 1-hour cap) is
    announced in Retry-After. Quota exhaustion on paid-door surfaces answers 402 with x402
    accepts[] rather than 429.
versioning:
  api: unversioned single surface; GET /api/health returns the deployed commit
  subregistry: /v0.1/servers pinned to Official MCP Registry spec v0.1
request_tracing:
  documented: none (Cloudflare cf-ray only, infrastructure-level)
field_slicing:
  note: llms-full.txt supports ?prefix=/api/ to return only one branch of the reference
idempotency:
  coverage: partial
  scope:
    - like_listing
    - delete_api_listings_by_id_like
  mechanism: natural (state-based), no Idempotency-Key header
  docs: https://agentalog.com/llms.txt
  note: >-
    The provider documents that repeat likes do not add up ("the counter counts people") and
    unlike is a stable inverse — replay-safe by design on the engagement surface. No
    idempotency-key mechanism exists for the paid writes (create_listing, contact,
    post_api_credito); their guard is sequencing instead: validation (400) and quota (429)
    run BEFORE the 402 charge, and a settled-payment/failed-write answers 502 with a
    transaction id for manual reconciliation.
reversibility:
  grade: documented
  writes:
    - operation: like_listing
      reversal: delete_api_listings_by_id_like
      window: unstated ("gives the point back to the public counter")
      docs: https://agentalog.com/llms.txt
    - operation: post_api_listings_by_id_comments
      reversal: delete_api_comments_by_id
      window: unstated (own comments only; someone else's answers 404)
      docs: https://agentalog.com/llms.txt
    - operation: create_listing
      reversal: patch_api_listings_by_id (edit; changing the URL re-queues) — no documented delete
      window: unstated
      docs: https://agentalog.com/llms-full.txt
    - operation: post_api_credito
      reversal: none documented — prepaid credit top-up (x402 on-chain USDC) has no published refund path
      window: none stated
      docs: https://agentalog.com/llms-full.txt
  note: >-
    Reversal paths exist and are documented for engagement writes, but no reversal WINDOW is
    stated anywhere, so the grade stays documented, not verified. x402 payments themselves
    are not documented as refundable.