generated: '2026-08-02'
method: searched
source: https://docs.aigateway.cequence.ai/docs/guides/security-policies
summary: >-
  Rate limits in the Cequence AI Gateway are configured per tool on an MCP server, on a rolling window.
  Administrators set Max Requests (1 to 1,000,000) and a Time Window (1 second to 24 hours). When rate
  limiting is enabled without explicit per-tool overrides, method-derived defaults apply.
model: per-tool, rolling-window
configurable:
  max_requests: {min: 1, max: 1000000}
  time_window: {min: 1 second, max: 24 hours}
rate_limits:
- name: Read methods default
  applies_to: GET, HEAD, OPTIONS
  limit_count: 1000
  limit_period: 1 hour
  scope: per tool
- name: Write methods default
  applies_to: POST, PUT, PATCH
  limit_count: 100
  limit_period: 1 hour
  scope: per tool
- name: Delete methods default
  applies_to: DELETE
  limit_count: 10
  limit_period: 1 hour
  scope: per tool
exceeded_behavior:
  http_status: 429
  status_text: Too Many Requests
  agent_message: Error message indicating the limit was exceeded.
  audit_reason: rate_limit_exceeded
  retry: Agent can retry after the time window resets.
headers:
  documented: false
  note: >-
    No RateLimit / X-RateLimit response header set is documented; the signal to the agent is the 429
    status plus the error message.
circuit_breaking:
  enabled: true
  behavior: >-
    Automatically halts traffic to an unresponsive upstream (503 response) when failures are detected,
    with automatic recovery testing.
related:
  errors: errors/cequence-security-problem-types.yml
  conventions: conventions/cequence-security-conventions.yml
x-evidence:
  fetched: '2026-08-02'
  urls:
  - url: https://docs.aigateway.cequence.ai/docs/guides/security-policies
    http_status: 200