generated: '2026-09-19'
method: searched
source: https://perkoon.com/.well-known/agent-card.json (rateLimits field)
docs:
- https://perkoon.com/.well-known/agent-card.json
- https://perkoon.com/llms.txt
- https://www.npmjs.com/package/@perkoon/mcp
- https://www.npmjs.com/package/perkoon-transfer
limit_count: 4
summary: >-
  Perkoon publishes its rate limits in four places that agree with each other, including inside the A2A agent card
  itself (a non-standard rateLimits field), which is the one place an agent reads before its first call. Limits are
  per IP over a sliding 60-second window and are bucketed by request type on the REST paths: 10 session creates, 30
  joins and 20 status checks per minute. On /a2a every JSON-RPC POST - send-files, receive-files and session-status
  alike - counts against the 10/minute create budget. Exhaustion returns HTTP 429 with a Retry-After header in
  seconds, mirrored as retry_after in the JSON body. Limits key on IP, not agent identity, so clients behind a shared
  egress (NAT, CI, cloud gateway) share one budget. The published Agent Skill adds an operational caveat: the CLI
  currently reports a 429 as exit code 3 (network/session error) before session_created, so a fast exit-3 after
  several rapid sends should be read as rate limiting. No limit was observed live: robots.txt disallows /api/ (honored)
  and /a2a answered a Cloudflare challenge, so nothing here is probed.
rate_limits:
- scope: per-ip
  surface: POST /api/v1/sessions (session create; CLI send, MCP send_file)
  window: 60s sliding
  limit: 10
  burst: null
  source: agent card rateLimits.createsPerMinute; llms.txt Rate Limits
- scope: per-ip
  surface: POST /api/v1/sessions/{code}/join (CLI receive, MCP receive_file)
  window: 60s sliding
  limit: 30
  burst: null
  source: agent card rateLimits.joinsPerMinute; llms.txt Rate Limits
- scope: per-ip
  surface: GET /api/v1/sessions/{code}/status (MCP check_session)
  window: 60s sliding
  limit: 20
  burst: null
  source: agent card rateLimits.statusChecksPerMinute; llms.txt Rate Limits
- scope: per-ip
  surface: POST /a2a (every JSON-RPC method - send-files, receive-files, session-status)
  window: 60s sliding
  limit: 10
  burst: null
  note: shares the create budget regardless of skill; the separate join/status budgets apply only to the REST paths
  source: agent card rateLimits.a2a
headers:
  documented:
  - {header: Retry-After, unit: seconds, on: 429}
  body_field: retry_after (same value as Retry-After)
  observed: []
  note: Not observed live (no request that could trigger a limit was made). No X-RateLimit-* / RateLimit-* quota headers are documented.
exhaustion_status: 429
client_guidance:
  verbatim: 'llms.txt: "Honor it — wait, then retry."; SKILL.md: "back off ~30s and retry, don''t hammer"'
  cli_mapping: 'a 429 surfaces as CLI exit_code 3 before session_created (SKILL.md, perkoon-transfer 1.1.0)'
  mcp_mapping: 'send_file / receive_file may return isError: true with a "Too many requests" message; transient, wait ~60s (README, @perkoon/mcp 0.3.0)'