Skip to main content

Agent Operation Patterns

Use these patterns when exposing Zerq MCP endpoints to AI agents that need deterministic behavior and tight scope.

Pattern 1: Discover then act​

Allow read-only discovery tools first, then permit mutating tools only for explicit workflows.

Pattern 2: Policy-first routing​

Map each agent identity to a dedicated client and profile. Do not share broad profiles across multiple agents.

Pattern 3: Bounded retries​

Agents should retry transient failures, but stop on policy responses (401, 403, 405) and require operator review.

Example tool call flow​

curl -i "https://gateway.example.com/mcp" \
  -H "Authorization: Bearer $AGENT_TOKEN" \
  -H "X-Client-ID: support-agent" \
  -H "X-Profile-ID: prod-ai-support" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":"tools-1","method":"tools/list","params":{}}'

Status handling for agents​

  • 401: refresh credentials or re-authenticate.
  • 403: tool is outside policy scope; do not auto-retry.
  • 405: wrong method for MCP endpoint; fix client implementation.
  • 429: back off with jitter and respect retry window.