AI agent connectivity

Your AI agent is ready for the enterprise. Its data is stuck behind a firewall.

Ship one lightweight ngrok agent per customer and your AI agents reach the internal APIs, databases, and MCP servers they need. Each customer stays isolated, and traffic is encrypted end to end—so you never become a subprocessor of their data.

Follow the quickstart Talk to an engineer

How it works

One ngrok agent per customer. Every system your AI agents need.

Your customer runs a lightweight ngrok agent behind their firewall. It dials out to ngrok and gives your AI agents private, scoped URLs to only the APIs, databases, and MCP servers you're authorized to reach.

Customers run a lightweight ngrok agent that creates secure tunnels: outbound TLS connections from ngrok agents to the ngrok cloud on port 443.

Authorize your AI agents' access with mTLS, IP restrictions, or JWTs. ngrok relays traffic from your product directly to the target service.

Reach every service your AI agents need over one ngrok agent: REST APIs, databases, internal web apps, and private MCP servers, each on its own scoped URL.

Expand from one to many customers with the same ngrok agent configuration and new endpoints with the layers of authentication they demand.

Why ngrok?

Unblock the deal that was stuck on access. Your customer runs one ngrok agent, not a networking project. Live in days, not quarters.

Keep customer networking out of your support queue. Your team focuses on your product's value instead of firewall tickets and peering issues.

One integration reaches every customer environment. AWS, Azure, on-prem, and hybrid environments all use the same connection model.

Databricks did it. You can too.

Read the case study

Proven at enterprise scale.

Databricks evaluated 17 alternatives before choosing ngrok for secure ingress into customer environments across cloud and hybrid deployments.

Cut customer deployment time from months to hours and days.
Reached enterprise preview in roughly three months.
Reduced security review cycles with a single ingress model.
Eliminated public IP requirements across worker nodes.

Security

Pass your customers' security reviews on the first try.

The ngrok agent already runs in banks, healthcare systems, and Fortune 100 networks. End-to-end TLS means the ngrok cloud only ever sees ciphertext, so you never become a subprocessor of their data.

Give your AI agents one service, not the whole network

Your customer exposes exactly one service by name—nothing else in their network is even addressable. Scope access to the APIs, databases, and MCP servers your AI agents need and not a single port more.

Diagram showing scoped access to a single service in a customer environment

Never become a subprocessor

Terminate TLS in your customer's network at the upstream service or the ngrok agent. The ngrok cloud service only sees ciphertext, so your AI agents' traffic stays private end to end.

Restrict traffic regions for data residency

Comply with data residency requirements by selecting the exact ngrok data centers used to relay your AI agents' connections.

Isolate every customer, credential by credential

Each customer gets its own scoped, endpoint-bound token and its own policy. A credential stolen at one customer can never touch another's endpoints, and offboarding is a single token revocation.

All the boxes you need to check

Your customers have questions.
We have answers.

Send your customers a complete Q&A on how ngrok works and why it's secure.

Check it out

Reliability

Hit all the 9s in your SLA.

Network failures are inevitable. Identify and recover from them automatically.

Reconnect automatically

You can't control your customer's network. That's why the ngrok agent runs in the background and heartbeats its connection to recover quickly after it sees connection reset by peer.

Heartbeat connection recovery visualization

Alert on issues before customers notice

Publish tunnel status and connection events to your telemetry platform. When a connection drops, you'll know before your customer does.

High availability with ngrok agent redundancy

Run multiple ngrok agents in your customer's network and ngrok will balance connections among them. You'll stay connected even when a machine running one ngrok agent fails.

Stay online during region outages

Agents create secure tunnels to multiple regions of the ngrok cloud service. You won't go down when entire datacenters fail (cough, us-east-1, cough).

Production Ready

Built to run a fleet of ngrok agents, not a tunnel.

Ship the ngrok agent to every customer

The ngrok agent is a cross-platform, dependency-free binary pre-packaged for Docker, Kubernetes, Windows, Linux, and macOS.

Package installation commands for Windows, Linux, Kubernetes, and Docker

`import ngrok`

Embed the ngrok agent directly in your product with an SDK in five languages—ideal when your own software already runs inside the customer environment.

SDK installation commands for Rust, Python, Node, and Go

Provision every customer with an API call

APIs for every feature, a Terraform provider for IaC, and bot users plus wildcard domains to automate per-customer onboarding as your ngrok agent footprint grows.

White-label everything

Brand your URLs, ngrok agent connect address, and dedicated IPs with your own domains so the customer's security team allowlists your product—not a generic ngrok endpoint.

Need to self-host ngrok?

Inquire about private edition

Connect your AI agents to your first customer network in 10 minutes.

No upfront costs. No contact sales. Pay only for what you use.

Follow the quickstart Talk to an engineer

Frequently asked questions

It's the pattern where your AI product needs to act on systems inside your customers' networks—calling internal APIs, querying databases, or reaching private MCP servers behind a firewall. ngrok gives each customer a lightweight ngrok agent that connects those systems to your product over a secure tunnel, so you don't need your customer to adopt a VPN, open firewall ports, or set up VPC peering.

Anything with a network address: REST APIs, databases (PostgreSQL, MySQL, MongoDB), internal web apps, legacy systems, and MCP servers. If it has a port, your AI agents can reach it over a secure tunnel—scoped to exactly that service and nothing else.

Yes. Run the ngrok agent next to the MCP server and your product gets a private, authenticated endpoint to it—no public exposure and no inbound ports on the customer's side. See our MCP guide for details.

A VPN gives your AI agents broad network access and requires your customer's IT team to configure and maintain it. ngrok scopes access to specific services by name, runs as a lightweight ngrok agent your customer deploys, and connects outbound over port 443—so nothing else in their network is even addressable to your AI agents.

No. With end-to-end TLS, you terminate encryption at the upstream service or the ngrok agent inside your customer's network, and the ngrok cloud only ever relays ciphertext. Your AI agents' traffic stays private end to end, which keeps you off your customer's subprocessor list.

Every customer site gets its own scoped, endpoint-bound token and its own traffic policy (IP allowlists, mTLS, JWT rules). A credential stolen at one customer can only ever touch that customer's endpoints, and offboarding a customer is a single token revocation.

ngrok's unit of exposure is a named service bound to one host and port, reachable only through an explicit policy, with application-layer controls (JWT, OAuth, mTLS, rate limits) on every request. Tailscale's unit is a network route: a subnet router advertises a CIDR into a shared network and an ACL subtracts what shouldn't be reachable. For a multi-tenant AI product, that additive, per-service model plus per-customer credential isolation is the difference. If your AI agents need true LAN presence or raw peer-to-peer throughput, Tailscale is a better fit. See our ngrok vs. Tailscale comparison.

Provision each customer programmatically: APIs for every feature, a Terraform provider for infrastructure as code, bot users, and wildcard domains. Run multiple redundant ngrok agents per customer for high availability, and embed the ngrok agent in your product with an SDK in five languages.

No. The ngrok agent makes outbound TLS connections to the ngrok cloud on port 443, the same port as regular HTTPS traffic, to establish the secure tunnel. Most firewalls allow this by default.

The ngrok agent heartbeats its connection to the ngrok cloud every 10 seconds. If it doesn't get a response within 15 seconds, it terminates and immediately attempts to reconnect. You can also run multiple redundant ngrok agents per customer, and ngrok agents connect to multiple regions so a regional outage doesn't take you down.

SOC 2 Type II, HIPAA/BAA, GDPR, CCPA, and EU-US DPF. Full documentation and audit reports are available at trust.ngrok.com.

You only pay for what you use. Pricing is based on how often your endpoints are active, the traffic you push through ngrok, the number of Traffic Policy executions, and more. See full pricing.

Your AI agent is ready for the enterprise. Its data is stuck behind a firewall.

One ngrok agent per customer. Every system your AI agents need.