Getting started with Cue
Cue is two things in one calm surface: a notification hub that gathers what matters into one inbox, and a human-in-the-loop waiting room where your AI agents block until you make a decision.
What Cue does
Most of what needs your attention is scattered — signups in one dashboard, payments in another, deploys in a third, and a row of terminals where AI agents quietly wait for a yes or no. Cue pulls all of it into a single, quiet inbox and reserves a single colour, amber, for the one thing that actually needs you.
- Notification hub — connect a project or service and Cue gathers what matters (signups, payments, deploys, errors) into one inbox.
- Human-in-the-loop for agents — agents connect over MCP; when one needs a decision it raises a request and blocks until you respond.
Core concepts
Three nouns carry the whole model. Keep streams and connectors distinct — a stream is the lane, a connector is the source that fills it.
An inbox lane. You sort cues into streams — one per project, service, or agent. A stream is just the lane; it does not fetch anything itself.
An integration that feeds cues INTO a stream — GitHub, Stripe, Vercel, Sentry, Linear, a raw webhook, or an MCP agent. The connector is the source; the stream is where its cues land.
One item in a stream. A notification is a quiet row you can clear or snooze. An agent request is a raised, blocking decision card — an AI agent waits on it until you approve or decline.
The amber law
Cue is near-monochrome on purpose. One accent colour — amber #F5A524 — is reserved exclusively for things that need your decision: a blocking agent request, the needs-you dot, the needs-you count. Cleared items, quiet notifications, and body text never wear it. When you see amber, something is waiting on a human.
Quickstart
Three steps from empty to a live inbox.
Create your workspace at app.attentioncue.com. Every account gets a private inbox and its own MCP endpoint.
Add a lane for the project, service, or agent you want to watch — for example "web-app" or "Billing". Cues land here.
Add a connector to feed the stream: point a service webhook at Cue, or connect an AI agent over MCP so it can raise decisions.
Next steps
- Connect an AI agent over MCP — give an agent your endpoint and let it block on
request_decisionuntil you approve or decline. - REST API reference — manage streams, read and clear cues, and decide agent requests programmatically.