• PASS
  • FAIL
  • HEALTH
  • HUMAN

How to audit authorization policies with your AI

Connect Cerbos to your AI client to run permission checks and monitor instance status. Your agent handles the heavy lifting of evaluating policy logic against specific scenarios.

Ask AI about this page

Short answer

Can my AI audit my Cerbos authorization policies?

Yes. Your AI uses the Cerbos MCP to run batch evaluations of your policies and pull real-time health data from your instance. You'll get a clear view of permission consistency and system status.

Where every issue lands.

Audit outcomes

The AI categorizes the results of your policy checks into actionable buckets.

PASS

Consistent permissions

The AI confirms that your policies behave exactly as expected for the tested subjects. You get a clean report of successful evaluations.

FAIL

Policy mismatches

The AI identifies where authorization logic fails to meet your requirements. It flags specific subjects or resources that triggered an unexpected result.

HEALTH

Instance status

The AI pulls telemetry to show how your Cerbos instance is performing. You'll see if the system is running smoothly or hitting bottlenecks.

HUMAN

Manual intervention

When the AI finds a logic gap it can't resolve, it flags the specific policy for you to review. You decide how to fix the underlying code.

The workflow

What your AI does when the request arrives.

Your agent acts as the bridge between your policy logic and your audit requirements.

  1. Check system health

    The AI first looks at the current state of your Cerbos setup to ensure the environment is stable for testing.

    get_server_info

  2. Run permission tests

    The agent executes a series of tests against your defined policies to see if they grant or deny access correctly.

    authzen_evaluations

  3. Identify discrepancies

    The AI compares the actual results of the evaluations against your intended security model.

    authzen_evaluations

  4. Report findings

    The agent summarizes the health of the instance and the results of the batch evaluations for your review.

    get_server_infoauthzen_evaluations

Try it

Copy these to start.

Use these prompts to kick off your audit.

Starting points

These are starting points. Swap out the specific policy names or resource IDs to match your setup.

Cerbos Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_y0c5dNo7ENj8oh5gsdkYK0hZAMSRgY44sro1JyED/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — Cerbos capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "cerbos-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_y0c5dNo7ENj8oh5gsdkYK0hZAMSRgY44sro1JyED/mcp"
    }
  }
}
Copy into chat04

  • Run a batch evaluation on my 'user_access' policy for the 'admin' role.

  • Check the health of my Cerbos instance and tell me if there are any issues.

  • Evaluate if the 'billing_service' policy correctly denies access to 'guest' users.

  • Run an audit on all policies related to the 'orders' resource.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect Cerbos once, then ask.

You only need to link your Cerbos instance once. After that, your AI client can run audits and health checks whenever you need them.

Connect Cerbos to your AI

FAQ

How this task behaves

  • 01

    Can the AI change my authorization policies?

    No. The AI can only read your policies and run evaluations. It cannot modify your Cerbos configuration or rewrite your policy files.

  • 02

    Does the AI delete any user data?

    No. The tools provided only allow for evaluating permissions and checking server info. There are no tools for data deletion.

  • 03

    How does the AI know if a policy is 'correct'?

    The AI compares the results of the evaluations against the specific scenarios or expected outcomes you provide in your prompt.

  • 04

    Can I use this to monitor my production instance?

    Yes, you can use the health check capability to monitor the status of your Cerbos instance.

  • 05

    What happens if an evaluation fails?

    The AI will report the failure and explain which subject or resource caused the mismatch based on the evaluation results.

  • More questions about Cerbos? See everything the Cerbos Connector can do

Connect Cerbos to Claude, Cursor, ChatGPT & more