A screen that AI agents can draw on
AI conversations are stuck in text. An agent that could compose charts, dashboards and diagrams on any nearby screen would show far more than it can say.
Display · Solo Side ProjectThe problem
An AI agent can analyse a spreadsheet or plan a week, but it hands back paragraphs. Display adds a live visual canvas to any AI conversation: connect an MCP-capable agent and it composes dashboards, charts and diagrams on any screen in real time, whether that's a TV, a tablet, a phone or a second monitor.
Constraints
- Agents aren't designers. Given free rein over pixels, a model produces inconsistent layouts. It needs a small vocabulary that always looks good.
- Screens differ. The same content has to adapt to a television across the room and a phone in the hand.
- Live, across devices. Several screens can show the same session, and each update has to land everywhere at once.
What I did
Display is a solo side project, and I built it end to end. The agent connects to a session-scoped MCP server, which exposes a set of composable, editorial render primitives rather than raw drawing commands. The agent decides what to show; the primitives decide how it looks, and adapt their layout to the surface they're shown on. Updates sync in real time across every connected screen over AWS AppSync Events.
Architecture
Outcome
Display is in production at display.sh, built and shipped by me alone.