Which flavor is for you?

OPC UA Modeler runs in three places, and the YAML you write is the same in all of them. What differs is where the work happens, what you install, and what each can do today.

Start here

You know OPC UA, you have never used us. Start online. There is nothing to install, the editor has full language support, and you will have a validated model in a few minutes. Move to the CLI when you want it in a build.

You are wiring up an AI agent. Start with the MCP server. It is on public npm and its type-catalogue tools work offline with no account, so you can explore the whole OPC UA companion-spec type system before deciding anything.

Your models cannot leave your network. Use the desktop CLI. Everything that matters — generate, reverse, validate, the language server — runs with no network at all.

You are evaluating. Read the table, then go online. It is the fastest way to see what the tooling actually produces.

At a glance

	Online	Desktop CLI	AI agents (MCP)
Install	nothing — a browser	npm, Sterfive registry	npx, public npm
Works offline	no	yes, fully	type catalogue only
What you need	an account (Early Bird)	a commercial licence	nothing, then an API key for generation
Write and validate YAML	✅	✅	✅ validate
Generate NodeSet2.xml	✅ via the REST API	✅	✅ needs a key
Reverse an existing nodeset	✅ via the REST API	✅	✅ needs a key
Generate a model from a description (AI)	✅	—	✅ needs a key
Use your own vendor NodeSet2 files	✅ upload	✅ files on disk	not yet
Run the model in a live OPC UA server	✅	✅ generate --server	—

Online — the browser app

opcua-modeler.sterfive.io ↗

Nothing to install. Sign in with Google, GitHub, or an email address, and you get the full editor: the same language server as the desktop IDE, so you get live diagnostics, completions and hover documentation as you type; auto-generated diagrams and a documentation preview beside your model; a problems panel; and the ability to upload your own vendor NodeSet2 files as dependencies.

Two things it can do that the desktop cannot:

• Describe a model in plain language and have it generated, validated, and auto-corrected for you.
• Run your model in a real OPC UA server with one click, and point a client at it.

You can also publish a model to the community gallery, or fork someone else's, and issue API keys for the REST API.

Access is currently Early Bird, for adopters working with us directly — talk to us and we will set you up.

Desktop — the command line

The CLI is the original product and still the most complete. It is a commercial licence, installed from Sterfive's own npm registry:

npm config set @sterfive:registry=https://npm-registry.sterfive.fr
npm install -g @sterfive/opcua-modeler

Everything runs on your machine: generate, reverse, symbols, bundle, the documentation viewer, and the language server. No network, no account, no model leaving your site. That is the point of it — and it is why the CLI is what you want in CI, in a regulated environment, or behind an air gap.

It also installs the editor integration for you:

opcua-modeler install-extension

which sets up VS Code (or Cursor, or VSCodium) with the extension and the YAML schema, pointing at the CLI you just installed.

If you want the API contract without the cloud, opcua-modeler serve exposes the same /api/v1 endpoints over a local socket — see the LSP and serve commands.

Licensing and purchase: sterfive.com/product/opcua-modeler ↗

AI agents — the MCP server

Setup guide · public npm, Apache-2.0

npx -y node-opcua-modeler-mcp-server

This is the lowest-friction way in. Seven of its twelve tools answer from a catalogue bundled with the package — offline, no account, no key — covering every companion specification we ship: list namespaces, list and search types, inspect a type's members, resolve dependency order, find a reusable interface or AddIn, and look up an official engineering unit.

Two more work anonymously over the network: fetching the DSL grammar reference, and validating a model. Only generation, reverse-engineering and AI model creation need an API key.

There are two ways to connect it:

• Remote connector — paste a URL into Claude, nothing installed. Note the three key-gated tools are unavailable this way, because the hosted connector cannot carry your key.
• Local (stdio) — run it yourself with OPCUA_MODELER_API_KEY set, which unlocks everything. It can also be pointed at a local opcua-modeler serve so that models never leave the host.

Using more than one

They are designed to combine, and most people end up doing so:

• Agent drafts, human reviews. Have an agent produce a first model through MCP, then open it in the browser editor or your IDE to refine it.
• Author anywhere, build in CI. Write online or in VS Code; run opcua-modeler generate --strict in your pipeline so a broken model fails the build.
• Explore hosted, ship on-premise. Use the online app to learn the DSL, then move to the CLI once the models are real and cannot leave your network.

The YAML is portable in every direction. What does not travel automatically are the vendor NodeSet2 files a model imports — those live on disk on-premise and in your account online. opcua-modeler bundle packages a model together with its dependencies so it compiles on a machine that has never seen them.