The Safety Layer for AI Agents

Control what AI agents can do, and what requires human approval.

Start Free Trial Request Demo https://calendar.app.google/FV95tXZtfGpPk7398 grid

Define safety policies

Allow safe operations, deny dangerous ones, require approval for the rest. Policies can be as fine-grained as you need, down to specific parameter values of MCP tools.

Full visibility into AI actions

OpenClaw, Claude Cowork, or any agent to Preloop. Acts as an MCP proxy. Every tool call is logged with full context. No SDK, no code changes required.

Approve from anywhere

When AI attempts an action requiring approval, get notified instantly. Approve or reject from your phone, watch, or email. Agents wait for your decision.

What is Preloop?

Preloop is a policy engine for AI agents. It lets you define exactly what AI can do (allow), what it cannot do (deny), and what requires human approval. Every action is logged for full visibility.

Why do I need policies for AI agents?

AI agents can deploy code, delete databases, spin up expensive cloud resources, and manage infrastructure. Without policies, one mistake can cost hours of downtime or thousands of dollars. Preloop lets you define guardrails.

How does it work?

Connect your AI coding agent (Claude Code, Cursor, Windsurf) to Preloop via MCP. Define policies for each tool. Actions matching allow policies execute immediately. Actions matching deny policies are blocked. Everything else can require approval.

What kinds of actions can I control?

Any operation exposed via MCP. Deployments, database changes, billing modifications, secret management, and more. Policies can be fine-grained, checking specific parameter values and context.

Which AI agents does Preloop support?

Preloop works with any MCP-compatible AI agent, including Claude Code, Cursor, Cline, and Windsurf. Setup takes about 3 minutes.

How do approval notifications work?

When an action requires approval, Preloop sends notifications via mobile app, Slack, Mattermost, or email. Approve or reject from anywhere. Agents wait for your decision before proceeding.

Will policies slow down my AI agents?

Only actions requiring approval pause for human input. Allowed actions run instantly. Denied actions fail immediately with a clear message. Most workflows run without any delay.

What happens when I reject an action?

The AI agent receives your rejection with an optional reason. It can then stop, ask you for guidance, or try a safer alternative approach.

Do I get an audit trail?

Yes. Every AI action is logged with full context: what was attempted, which policy matched, who approved or rejected it, and when. Essential for security reviews and compliance.

Set up AI safety policies in 3 minutes Get Started with Preloop MCP 
claude mcp add --transport http preloop https://preloop.ai/mcp/v1
  --header "Authorization: Bearer YOUR_PRELOOP_API_KEY"

# Create or edit ~/.cursor/mcp.json
{
  "mcpServers": {
    "preloop": {
      "transport": "http",
      "url": "https://preloop.ai/mcp/v1",
      "headers": {
        "Authorization": "Bearer YOUR_PRELOOP_API_KEY"
      }
    }
  }
}

{
  "mcpServers": {
    "preloop": {
        "transport": "http",
        "url": "https://preloop.ai/mcp/v1",
        "headers": {
            "Authorization": "Bearer YOUR_PRELOOP_API_KEY"
        }
    }
  }
}