What can an AI assistant do with Szenera?
What the MCP server offers, what it leaves out on purpose, and why nothing reaches a display without your confirmation.
- For
- Anyone who wants to decide whether to let an assistant work with their displays, and developers who connect one.
On this page
An assistant connected through the MCP server works with the same inventory as you in the portal: it lists displays, reads scenes and programs, prepares drafts, and changes what a display shows. It does nothing your account can't do, and nothing that changes a display happens without your explicit confirmation.
What the server is
The MCP server is a thin adapter in front of the same API that the portal and the command line use. It holds no data and has no access of its own. Your credential travels with every request to the platform, and the platform decides what is allowed. Every rule that applies in the API applies here too, and there is no route or field that exists only here.
It adds four things on top of the API:
- A selection. Not every route becomes a tool, only those an assistant needs for its tasks.
- A confirmation before every change.
- Limits for the size of results and the runtime of a call.
- One narrow exception for deleting: the only thing a tool can remove is a display that runs in a browser.
Three kinds of tools
| Kind | What it does | Examples |
|---|---|---|
| Read | Looks at the inventory and changes nothing | Which displays are offline, what a scene contains, what a display shows right now as an image, what could be cleaned up |
| Propose | Creates or changes drafts that appear on no display | A scene from a template, a new scene document, a playlist, a preview image of a published version |
| Write | Changes what displays show or how they are set up, always with confirmation | Publish a scene, assign a program, override a display for an hour, restart a device, create a location or a web display |
The complete list with arguments and permissions is under All tools.
Nothing happens without your confirmation
Every writing tool first answers with a preview: which displays it would affect, what they are called, whether they are online, and what would be replaced. The assistant shows you this preview. Only when it calls the same tool again with the same values and confirm: true does the change happen.
The preview holds nothing in reserve. If something changes in between, the API's check for concurrent changes applies. A repeated call with the same intent takes effect once, because the server derives its idempotency key from the session, the tool, and the arguments.
What an assistant can't do
- Accounts, roles, and API keys. Not wired into the server at all.
- Billing and packages. Not available.
- Platform operations. Nothing the operator of the platform does.
- Deleting. No scene, media item, playlist, or account. The one exception is a web display, which an assistant can create in a second and should be able to clean up again. A display with a device behind it is rejected.
How it treats your data
The server speaks English: tool names, fields, status values, and hints are English, whatever language you write in. Names and texts from your inventory, such as the title of a scene, are data for the assistant and never instructions. Every tool description says so, so that a scene called "Delete everything" is just a name.
Results contain no signed download addresses. An image of a display or a scene travels as an image in the answer, not as a link that would end up in logs.
Was this helpful?