Manage Your Entire Edge From Your AI Assistant

Today the Ngris Model Context Protocol (MCP) server is live at mcp.ngris.com. Connect it to Claude Code, Claude Desktop, or any MCP-capable AI client, and you can drive your entire Ngris edge in natural language — expose a local service, deploy a site, inspect and replay a failing request, tighten a firewall rule — by asking your assistant, with no dashboard clicks in between.

Ngris has always been two doors onto one policy edge: expose a localhost or Kubernetes service through a tunnel, or deploy a site or app to that same edge. Both sit behind the same per-endpoint policy engine — a WAF, edge auth, rate limits, mTLS. The MCP server puts all of that behind a conversation.

Why AI-native edge control

Most edge and tunnel tooling assumes a human at a dashboard or a hand-written config file. But more and more of the work already happens next to an AI assistant: you scaffold a service in Claude Code, you paste a stack trace, you ask it to write the deploy step. The moment you need a public URL or an edge rule, you leave that context, open a browser, click around, and come back.

The MCP server removes that round trip. Your assistant already knows the shape of your app; now it can act on the edge too — the same edge that serves your production traffic — through Ngris' public, user-facing API. As far as we can tell, no other tunnel of this kind ships an MCP server: ngrok, Cloudflare Tunnel, and Tailscale Funnel don't expose one. That's the difference — Ngris is AI-native, so you manage the whole edge from your AI assistant.

Connect in about 30 seconds

In Claude Code, add the server over Streamable HTTP and authenticate:

# Add the Ngris MCP server (Streamable HTTP)
claude mcp add --transport http ngris https://mcp.ngris.com/mcp

# Then authenticate — a browser tab opens; click Authenticate, approve

Two ways to authenticate, both first-class:

• OAuth 2.1 (browser login) — your MCP client opens a login tab, you approve, and it holds the token. Nothing to copy or paste. Ngris' own account service is the authorization server; the MCP server is a resource server that verifies your token.
• An Ngris API key — for scripts, CI, or clients you'd rather configure with an environment variable. Create one in your account and pass it as a bearer credential.

Other MCP clients (Claude Desktop, editors, agents) connect the same way — Streamable HTTP at https://mcp.ngris.com/mcp for the hosted server, or stdio if you're running it locally.

What you can ask for

The server exposes roughly 150 tools, organized around the same surfaces you'd use in the dashboard. In practice you don't call tools by name — you describe the outcome and your assistant picks them. A few real flows:

• Expose a local port. "Give me a public URL for the app on port 3000." The assistant creates the endpoint and hands back the URL.
• Deploy a site or app. Ship a static build or a managed backend, list deploys, tail build logs, set a preview, or roll back to the previous release.
• Configure endpoints. Routing rules, rate limits, firewall rules, mTLS, per-endpoint auth and RBAC, health checks, custom error pages, and alerts — created and adjusted in place.
• Manage domains, SSL, and DNS. Check availability, add a custom domain, verify a CNAME, reissue a certificate, manage DNS records.
• Inspect and replay live traffic. Pull recent requests, search them in plain language, get an explanation of one, and replay the exact failing request.
• Run the account. Members and roles, SSO config, API keys, MFA status, audit logs, usage and billing — plus support tickets and the kanban board.

A concrete debugging loop: ask your assistant to find the 5xx responses on an endpoint, explain the worst one, and replay that request against your fix — all in the same conversation where you wrote the fix.

Preview before you change anything

Edge rules are load-bearing, and an AI proposing a change is not the same as an AI applying it blindly. The mutating policy tools pair with dry-run affordances: you can test a CEL expression and preview a rule's blast radius before it's saved, so your assistant can show you the impact first. That mirrors how Iris — the AI assistant built into the Ngris dashboard — already gates writes behind a confirmation card. Read-only tools are marked as such; write tools declare whether they're destructive.

A note on scope and least privilege

The MCP server is deliberately a thin client over the public API. It has no database and no service-level token of its own — every tool forwards your credential to the Ngris API, which remains the single authority on what you're allowed to do. The practical consequence: the server can never do more than you could do yourself in the dashboard.

Two more boundaries worth stating plainly:

• User-facing scope only. No admin routes are reachable through the server — it's guarded structurally and defensively.
• Secrets stay one-way. Setting an environment variable is write-only; creating an API key returns the key once; billing tools never touch card data (checkout returns a Stripe client secret for a client-side confirm, and the billing portal returns a Stripe-hosted URL).

An account API key is bound to a single account. The few multi-account tools require an OAuth or session credential and say so in their descriptions.

Try it

If you already have an Ngris account, add the server and ask your assistant to expose something or deploy a small site. If you don't, you can start free — no card needed — and be through a working tunnel or a deployed site in a few minutes.

# Install the CLI
curl -fsSL https://ngris.com/install.sh | sh

# Add the MCP server to Claude Code
claude mcp add --transport http ngris https://mcp.ngris.com/mcp