Decide whether this agent may call this tool now

Verify Gateway is the policy decision point for agent-tool execution. It combines trust score, server policy, tool risk, payload classification, cost, user mandate, and approval context into one allow, deny, or approval-required decision.

Choose

/v1/route

/v1/route chooses the best server/tool path for a task before an attempted call exists.

Decide

/v1/decide

Decide means allow, deny, or require approval for a specific attempted call by a specific agent and user.

MCP tool

decide_agent_call

Agents can ask Verify for runtime permission through the MCP surface; the decision is recorded in Ledger.

Decision example

Request

{
  "agent_id": "agent_123",
  "user_id": "user_456",
  "server_id": "io.sentinelsignal/scoring",
  "tool": "score_workflow",
  "action": "invoke",
  "mandate_id": "mandate_abc",
  "payload_classification": [
    "phi"
  ],
  "requested_cost": 0.02
}

Response

{
  "decision": "allow",
  "reason": "server policy passed; tool approved; user mandate valid",
  "max_runtime_seconds": 30,
  "audit_required": true,
  "ledger": {
    "invocation_id": "inv_123",
    "evidence_pack_url": "/v1/ledger/evidence-pack?server_id=io.sentinelsignal/scoring"
  }
}

Why it matters

Runtime control

Static scores become enforceable runtime decisions tied to actual agent context.

Approval gates

Write-like tools, external sends, sensitive payloads, and degraded servers can require explicit approval.

Evidence trail

Every decision can carry a policy snapshot, risk snapshot, mandate evidence, cost, and ledger event hash.