Skip to main content
Send these prompts to an AI agent that’s connected to the ngrok MCP server. The AI agent answers from your own account, so every answer is about your endpoints, traffic, and team. Prompts that change your account, such as saving a Traffic Policy, ask you to approve the change first. See Make changes safely.

Try it now

Name a service in the second prompt to choose which one the AI agent looks at. In the third, ask for any response a custom-response action can send, such as JSON or an HTML page. To keep changing that page, see Build a page, then change it.

Understand what’s running

These prompts cover what’s connected right now, plus your reserved domains and certificates. Swap in the error code you got. If the ngrok error page also shows a request ID, paste both.

Analyze traffic

These prompts read Traffic Inspector, including TCP connections. How far back your AI agent can look depends on your plan’s retention. Narrow any prompt to a host, path, status code, method, or client IP, and ask for counts by hour, by day, or by country. Say 500 for HTTP 500 only, or 5xx for every server error, and name a time zone when you ask for counts by hour or day.

Create traffic rules

The AI agent writes Traffic Policy, checks it with the validator, and saves it to a Cloud Endpoint. The MCP server can’t write to your ngrok agent’s config file, but a coding agent can edit that file itself. Swap in your own paths, limits, and countries, or another OAuth provider, such as GitHub or Microsoft. See Traffic Policy actions for everything a rule can do. When you name a path such as /login, say whether you mean only that path or every path under /login/ too. An IP policy can apply to several endpoints, so a rule you add to one blocks the IP on all of them. To block an IP on one endpoint only, say which endpoint.

Set up endpoints and resources

The AI agent can create and change Cloud Endpoints, reserved domains, TCP addresses, vaults, and secrets. Swap in your own domain, and name the service or vault you mean. It asks before it deletes anything, and it can confirm that a secret exists but can’t read it back. A reserved address with no online endpoint may still belong to an ngrok agent that’s offline right now, so check each one before you approve a release.

Check your account

Change the time window, or ask about authtokens, SSH keys, or personal access tokens instead of API keys. The AI agent sees each credential’s description, owner, and dates, but never the key or token itself.

Analyze AI Gateway usage

These prompts use the AI Gateway tools. Request details and captured bodies need request capture turned on.

Send feedback

The AI agent sends feedback about the MCP server’s tools to the ngrok team. In the second prompt, replace <YOUR_IDEA_HERE> with your idea.

Chain prompts

Send the prompts in a chain one after another in the same session. Each prompt builds on the answer before it.

Build a page, then change it

Find bad traffic, then block it

Swap the login page for any path you want to protect, and change the second prompt to match.