Drop Alerts

Drop Alert
Your agents ask. You decide.

Your AI agents run without you. Right up until the moment one of them shouldn't.

Drop Alert is the phone on the other end of that moment. An agent hits a decision
it may not take alone, calls one tool, and stops. Your phone rings. You answer.
The agent carries on with your answer in hand.

CONNECT AN AGENT OVER MCP

Point your runtime at the Drop Alert MCP endpoint and it gets one tool:

ask_human

It takes a title, a question, an urgency, the options you want offered, and a
timeout. It blocks until you answer or the timeout runs out.

That is the whole integration. No SDK to install, no callback URL to host, no
polling loop to write. Anything that speaks MCP — a coding agent, a research
agent, an orchestrator, your own runtime — can ask a person a question and wait
for the answer.

OR SEND AN EVENT OVER THE API

Not everything is a question. One POST sends anything worth knowing about, fire
and forget, nothing blocks:

A deploy finished.
A payment landed.
A job failed at 3am.
A queue is backing up.
A threshold was crossed.

Requests and events sit in the same inbox and are kept apart where it matters:
what is blocking an agent is at the top under "Needs attention", and what is
merely worth knowing sits under it.

ANSWER IN ONE TAP, OR IN YOUR OWN WORDS

A request arrives with the buttons its agent offered — Approve, Reject, whichever
options it passed — and a text box when the agent will take free text. Tap one, or
type. Either way the answer goes back to the runtime that is waiting on it.

Every request carries its context: how urgent the agent said it was, which channel
it came from, when it was sent, when it expires, whether it needs acknowledging.
You are not approving a deploy from a title alone.

URGENT WORK RINGS LIKE A PHONE CALL

An agent blocked on something urgent does not arrive as a notification you find an
hour later. It takes the screen, names what it is waiting on, and offers two
buttons.

And it can only do that because you let it. Urgency is a request, not a command:
the agent says how urgent it thinks something is, you decide what that channel is
allowed to do to your phone, and your own quiet hours apply on top. An agent
cannot escalate by claiming to be important.

SEE WHAT IS CONNECTED

The Agents tab splits what you have connected in two: the runtimes that ask you
questions, and the services that only send events. Both are drawn from what each
one has actually done, not from what it claims to be. The connection card shows
your MCP endpoint, your events endpoint, and how many items are open right now.

WHO THIS IS FOR

Anyone running agents that shouldn't have the last word.

Developers whose coding agent needs a deploy approved.
Teams whose automations need a human decision in the loop.
Operators who want to be told, not to go and look.
Anyone with a script, a pipeline or a workflow worth hearing from.

STOP WATCHING THE RUN

Don't tail the log waiting for the question.
Don't keep the dashboard open in case something breaks.
Don't discover on Monday that an agent stopped on Friday.

Let it ask.

Drop Alert
Your agents ask. You decide.