Skip to main content

Security Governance

Govern MCP access in Zerq with strict identity segmentation, explicit tool scope, and audit-first change control.

Governance model​

  • One agent type -> one client ID -> one profile.
  • Approve tool catalog access per profile, not globally.
  • Require change review for new mutating tools.
  • Enforce quotas to prevent runaway agent loops.

Policy outcomes to monitor​

  • 401: bad or expired agent credential.
  • 403: agent attempted tool outside approved scope.
  • 405: tool endpoint called with unsupported method.
  • 429: agent exceeded configured throughput budget.

Practical verification​

# Allowed read tool
curl -i "https://gateway.example.com/mcp/tools/get-status" \
  -H "Authorization: Bearer $AGENT_TOKEN" \
  -H "X-Client-ID: ai-ops-agent" \
  -H "X-Profile-ID: prod-ai-ops"

Run one explicit denied test on each policy release to confirm 403 behavior remains intact.

Audit requirements​

  • Keep approver, requester, and reason for each policy/catalog change.
  • Retain tool invocation logs with request IDs for investigations.
  • Review denied-access trends weekly to detect drift or misuse.