Manage your entire edge from your AI assistant
Ngris runs a hosted MCP server at mcp.ngris.com. Connect Claude Code, Claude Desktop, or any MCP client and drive your ngris account in plain language — expose ports, deploy sites, inspect & replay traffic, configure edge policy. 150+ user-facing tools: near-full control of your account, from the assistant you already have open.
- Reads free, writes confirmed
- User-facing scope only
- OAuth or API key
Your assistant already knows your app. Now it can run your edge.
You scaffold a service in Claude Code, paste a stack trace, 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 — the AI you're already in can act on ngris directly, through the same public API you'd use by hand.
Any MCP client
Claude Code, Claude Desktop, and other MCP-capable assistants connect over Streamable HTTP at mcp.ngris.com/mcp — or stdio for a local process.
150+ user-facing tools
Endpoints, hosting, domains, edge policy, traffic inspection, account, billing, credentials, monitoring, support & boards — mapped one-to-one onto the ngris API.
Can't exceed you
Every tool forwards your own credential. No admin routes, no credential minting from an OAuth session — it can only do what you could do yourself.
Near-full control, grouped by domain
The tools map onto the same surfaces you use in the dashboard. Reads run freely; writes are confirmed before they run.
Endpoints
List, create & delete endpoints; health checks; the auth gate (none / basic / token / OAuth / JWT, force-HTTPS); client access; mTLS; OAuth providers; cache stats & purge; TLS cert assignment.
Applications & hosting
Create & deploy — upload a zip, one-shot quick-deploy to a live URL, or git-deploy — redeploy, roll back, read build logs, set env vars, get preview URLs, and provision a managed database.
Domains · DNS · TLS
Add, check & verify domains; certificate list, status & reissue; and DNS records — managed end to end from chat.
Edge policy & security
Firewall policies, rate limits, routing rules, and traffic policies — with a blast-radius preview and CEL test-expression before apply — plus custom error pages.
Traffic inspection
Logs, metrics (requests, bandwidth, errors), security analytics, smart search, analyze, export — and replay-with-edit: resend a captured request against your fix.
Account & members
Status, usage-vs-limits, settings, and the audit log; switch, create or delete accounts; members & RBAC; verification and KYC.
Billing
Subscription, invoices, and the billing portal; upgrade & preview a plan change; plans, coupons, add-ons, and dedicated IPs.
Credentials & auth
API keys, auth tokens, MFA, SSO / OIDC, and client CAs for mTLS — reads and confirmed writes.
Monitoring
Create and manage alert rules and test a notification channel — a threshold you describe becomes a live alert.
Support · boards · Iris
Open & search support tickets and the KB; product boards (create, move & comment on cards); and Iris nudges and onboarding.
Scope is user-facing only — no admin operations, and an OAuth/MCP session can't mint long-lived credentials. It never exceeds what you could do yourself.
Connect in 30 seconds
One command in Claude Code, then a browser approval. Other MCP clients connect the same way.
Add the server
In Claude Code: claude mcp add --transport http ngris https://mcp.ngris.com/mcp. Any MCP client can point at the same Streamable HTTP URL.
Authenticate
With OAuth 2.1 a browser tab opens — click Authenticate, approve, done. Nothing to copy or paste. Or pass an ngris API key as a bearer token for scripts and stdio.
Just ask
Describe the outcome — “expose port 3000”, “deploy this folder”, “rate-limit /api to 100/min” — and your assistant picks the right tools. Writes are confirmed first.
Safe to hand to an agent
The server acts on real infrastructure — so it's a thin client over the public API, bound to your credential, and unable to escalate.
User-facing scope only
No /admin/* route is reachable through the server — guarded structurally and defensively. It exposes only the account operations you can perform yourself.
Bound to your credential
There's no database and no service-level token. Every tool forwards your OAuth token or API key to the ngris API, which stays the single authority on what you're allowed to do.
Reads free, writes confirmed
Read tools run freely; write tools declare when they're destructive and are confirmed before they run. Mutating policy tools preview their blast radius first.
No credential minting
An OAuth/MCP session can't create long-lived API keys or agent tokens — the server refuses and points you to the dashboard. A connected agent can operate your edge but can't forge a durable key.
Two ways to bring AI to your edge
Iris is the AI inside the ngris dashboard. The MCP server brings the AI you already work in. Same honest guardrails; different place to stand.
| Iris (in the dashboard) | MCP server (your AI client) |
|---|---|
| Drafts edge rules from plain English, gated by a confirmation card | Runs 150+ tools across your whole account from Claude or any MCP client |
| Lives in the ngris dashboard & forms | Lives in the AI assistant you already work in |
| Best for tuning policy while you're in the product | Best for staying in your editor / agent workflow |
Frequently asked questions
What is the Ngris MCP server?
It is a hosted Model Context Protocol (MCP) server at mcp.ngris.com that lets AI assistants — Claude Code, Claude Desktop, or any MCP client — drive your ngris account in natural language. It exposes 150+ user-facing tools mapped onto the public Ngris API, so your assistant can expose services, deploy apps, configure edge policy, and inspect traffic on your behalf.
How do I connect it in Claude Code?
Run: claude mcp add --transport http ngris https://mcp.ngris.com/mcp, then authenticate in the browser. Claude opens the Ngris login, you approve, and the connection is bound — there is no key to copy or paste. Other MCP clients point at the same Streamable HTTP URL, or run the server over stdio locally.
Is it safe to give an AI assistant access?
The server is a thin client over the public API with no admin surface and no service-level token of its own. Every tool forwards your own credential, so the assistant can never do more than you could yourself. Reads run freely, writes are confirmed before they run, and an OAuth/MCP session cannot mint long-lived credentials.
How does it differ from Iris?
Iris is the AI built into the Ngris dashboard, drafting edge rules with a confirmation card and blast-radius dry-run. The MCP server brings the AI you already work in — it runs 150+ tools across your whole account from your own client. Same honest guardrails; you choose where to stand.
Run your edge from your AI assistant.
Start free — no card needed — connect the MCP server, and expose or deploy something by just asking.