Integrating Agents

Agent Lifecycle

1Register agentDashboard: Agents → Add Agent

2Copy API keyShown once after creation — add to agent config

3Add MCP serverDashboard: agent page → MCP Servers → Add Server

4Write policiesDashboard: Policies → New Policy

5Route trafficAgent connects to proxy endpoint

6Monitor & auditDashboard: Overview, Audit Trail, Agent Radar

In the Dashboard

1Go to Agents in the left sidebar and click Add Agent.
2Enter a name, optional description, and choose a risk tier (low / medium / high / critical).
3Click Create. Copy the API key shown — it is displayed only once.
4On the agent detail page, go to MCP Servers → Add Server. Enter the upstream URL and a label.
5Optionally toggle Shadow Mode on the binding to observe traffic before enforcing policies.

Via REST API

Register an Agent

SDK examples have been retired. For a supported integration, use the OpenAPI reference.

Response

json

{
  "id": "agt_01HXYZ...",
  "name": "coding-agent",
  "tenantId": "ten_01HXYZ...",
  "riskTier": "medium",
  "agentKey": "shld_agent_<token>"  // ← send this to your agent
}

Create an MCP Binding

Bindings map an agent to an upstream MCP server. You can have multiple bindings per agent (e.g., filesystem, database, web-search).

SDK examples have been retired. For a supported integration, use the OpenAPI reference.

Shadow Mode

Set shadowMode: true to log all traffic from this binding without applying enforcement. Useful for observing agent behavior before writing policies.

Multi-Agent Setup

Register multiple agents per tenant — each with its own key, risk tier, and bindings. Policies can be scoped per-agent or tenant-wide.

SDK examples have been retired. For a supported integration, use the OpenAPI reference.

Credential Rotation

Rotate an agent's API key without downtime. The old key remains valid for a brief grace period after rotation to allow graceful handover.

SDK examples have been retired. For a supported integration, use the OpenAPI reference.

Risk Tiers

Tier	Use Case	Enforcement
low	Read-only, trusted internal agents	Lenient — monitor only
medium	Standard development agents	Standard policy evaluation
high	Agents with write/execute access	Strict + human review triggers
critical	Prod infra, financial, PII access	Auto-block anomalies, mandatory review