Why built-in AI is a risk on screens
A model inside a signage platform is a new system with access to your content and, often, a route to every public screen you run. It is run by the vendor, usually through a model provider your security team never assessed, and it is switched on for everyone in the workspace. It is also rarely the best tool for the job: like the template editors before it, an assistant built into one product falls behind the ones your team already uses. The risk grows and the benefit shrinks at the same time.
What Screenly does instead
The platform stays free of AI, and it is open to the assistant you choose, through the same controls a person goes through.
- 01
Nothing is sent to a model
Screenly does not send your content, screens or workspace to an AI model. No assistant is switched on by default, and no model reads your playlists in the background.
- 02
Your assistant, not ours
Applies to: Claude, Cursor and other MCP clients
An assistant that supports the Model Context Protocol connects through the Screenly MCP server, which runs on your own computer as part of the Screenly CLI. The assistant in use is the one your organization already reviewed, under its own terms with its provider.
What the MCP server is - 03
Access set by a token
An assistant holds no access of its own. It gets exactly what its API token allows, in one workspace. Give it a token from an account with only the access it needs, and deleting the token stops it at once.
Permissions - 04
Screens stay with people
Screens are read-only to an assistant. It can report that a screen is offline and what it plays, but it cannot pair, change, reboot or delete one, so the hardware stays with the people who manage it.
- 05
Destructive actions are marked
Every action an assistant can take is marked read-only, change or destructive, so an assistant that supports confirmations asks before it deletes a playlist rather than after. Changes scoped to a label reach only that group of screens.
Working with an assistant - 06
The token stays in the keychain
Applies to: Claude Desktop extension
With the Claude Desktop extension, the token is kept in your operating system's keychain rather than in a configuration file.
AI assistants in Screenly
What to ask about AI in a signage platform
The questions a security review should put to any vendor that says its platform is AI-powered, and Screenly's answer to each.
| Ask | Screenly |
|---|---|
| Is our content sent to a model, and whose? | No. Nothing in the workspace is sent to a model. |
| Can we use the assistant we already approved? | Yes, any assistant that supports MCP |
| What can an assistant change without a person? | Only what its token allows, in one workspace; screens are read-only |
| Can a change be kept to some screens? | Yes, by scoping it to a label |
| How do we stop an assistant? | Delete its token |
Questions
Does Screenly use AI?
No. Screenly has no built-in AI and does not send your content, screens or workspace to an AI model.
Can I manage Screenly with ChatGPT, Claude or another assistant?
Any assistant that supports the Model Context Protocol can connect through the Screenly MCP server, which runs on your own computer. Claude Desktop installs it as an extension; Cursor and other clients use a short configuration.
Is it safe to let an AI assistant manage digital signage screens?
It is as safe as the access you give it. In Screenly an assistant has exactly the access of its API token, in one workspace, cannot pair, change, reboot or delete screens, and stops the moment you delete the token.
Why not build AI into the product?
A built-in model is one more system with access to your content and your public screens, run by us and a model provider you did not choose. Keeping it out leaves your team free to use the assistant it already trusts, under controls you set.