n8n-nodes-agentbouncr

n8n community node for AgentBouncr — the AI Agent Governance Platform.

Evaluate tool calls against governance policies, manage AI agents, control the kill-switch, and receive real-time governance events — all from your n8n workflows.

Installation

1. Open your n8n instance
2. Go to Settings > Community Nodes
3. Click Install a community node
4. Enter: n8n-nodes-agentbouncr
5. Click Install

Manual Installation

cd ~/.n8n/nodes
npm install n8n-nodes-agentbouncr

Restart n8n after installation.

n8n v2 Task Runner Compatibility

n8n v2 uses an external Task Runner that does not load community nodes by default. If you see "Unrecognized node type" errors after installation, set:

N8N_RUNNERS_DISABLED=true

This forces n8n to run workflows in-process where community nodes are loaded. This affects all community nodes, not just this one.

Docker: Add to your docker-compose.yml:

environment:
  - N8N_RUNNERS_DISABLED=true

Authentication

1. Sign up at agentbouncr.com
2. Create an agent token in the Dashboard (Agent Detail > Create Token)
3. In n8n, go to Credentials > Add Credential > AgentBouncr API
4. Paste your gov_ API key
5. (Optional) Change the API URL for self-hosted instances

Nodes

AgentBouncr

The main node for interacting with the AgentBouncr API.

Resource: Evaluate

Execute — Check if an AI agent is allowed to use a specific tool before execution.

| Field | Required | Description | |-------|----------|-------------| | Agent ID | Yes | The agent making the tool call. Auto-registered if unknown. | | Tool | Yes | The tool name (e.g. send_email, search) | | Parameters (JSON) | No | Tool call parameters (logged in audit trail) | | Trace ID | No | Correlation ID across systems |

Output:

{
  "allowed": true,
  "traceId": "abc-123",
  "reason": "Matched rule: allow-all"
}

Use an IF node after AgentBouncr to branch on allowed equals true.

Resource: Agent

| Operation | Description | |-----------|-------------| | List | Get all registered AI agents | | Get | Get details of a specific agent | | Create | Register a new agent with name, description, allowed tools, and policy | | Update | Change agent status (active/stopped/suspended/decommissioned) or policy binding | | Delete | Deregister an agent |

Resource: Kill Switch

| Operation | Description | |-----------|-------------| | Activate | Emergency stop — immediately block all AI agent actions | | Deactivate | Resume normal agent operations | | Get Status | Check if the kill switch is currently active |

AgentBouncr Trigger

Webhook-based trigger that fires when governance events occur. When the workflow is activated, it automatically creates a webhook subscription in AgentBouncr. When deactivated, the subscription is removed.

Available Events:

• tool_call.allowed, tool_call.denied, tool_call.error
• approval.requested, approval.granted, approval.rejected, approval.timeout
• agent.started, agent.stopped, agent.error, agent.suspended, agent.idle, agent.decommissioned, agent.config_changed
• killswitch.activated
• injection.detected, audit.integrity_violation
• policy.created, policy.updated, policy.deleted
• rate_limit.exceeded, license.validated, license.expired
• * (all events)

Default: tool_call.denied + killswitch.activated

Optional: Filter events by a specific Agent ID.

Example Workflows

Pre-Tool-Check Pattern

[Trigger] → [AgentBouncr: Evaluate] → [IF: allowed = true] → [Your Tool Node]
                                       └─[Denied] → [Slack: Notify Team]

Alert on Denied Tool Calls

[AgentBouncr Trigger: tool_call.denied] → [Slack: Post Message]

Emergency Stop via Manual Trigger

[Manual Trigger] → [AgentBouncr: Kill Switch Activate]

Auto-Registration

AgentBouncr supports auto-registration: if you use an Agent ID that doesn't exist yet in the Evaluate operation, the agent is automatically created with a default policy. No need to create agents manually first.

This is the "Stripe model" — your first API call is the registration moment.

Security Note

In v0.1.0, incoming webhook events from the AgentBouncr Trigger are not cryptographically verified (no HMAC signature check). HMAC verification will be added in v0.2.0. The webhook URL itself provides basic security as it is unique and not guessable.

Roadmap (v0.2.0)

• HMAC webhook signature verification on incoming trigger events
• Custom error messages (401, 403, 429 mapped to user-friendly hints)
• Approval operations (List, Approve, Reject)

Compatibility

• n8n: v1.0.0 and later
• Node.js: 20+

Links

• AgentBouncr Website
• AgentBouncr Dashboard
• npm Package
• GitHub Repository

License

MIT