How-to · Claude Code

Get a push when Claude Code needs a decision

Q: How do I get notified when Claude Code needs permission?

Two routes. DIY: wire ntfy into Claude Code's Notification hook so it fires a push when the agent stops or asks. Meltbox: your agent pushes a brief, and with Meltbox installed to your home screen your phone gets a push you can answer from the lock screen.

Q: Does it work when the app is closed?

Yes. Meltbox is an installed PWA with real push on Mac and iPhone, including when the app is closed.

Q: Do I need to install an SDK?

No. You install nothing on your side beyond adding Meltbox to your home screen. The agent redeems a one-time invite link and pushes briefs; you triage them.

Claude Code doesn't buzz your phone when it finishes or asks for permission. You tab away, lose the thread, and burn fifteen minutes getting back in. Here's the honest DIY fix — and the version where the notification is the decision.

Create a free workspace See the live demo

Short answer. Two routes. DIY: wire ntfy into Claude Code's Notification hook so a push fires when the agent stops or asks permission. Meltbox: your agent pushes a brief; with Meltbox added to your home screen you get a push you can answer from the lock screen — the choice flows back as the agent's next step.

"Doesn't Claude Code already notify me?"

It can — a terminal bell, an iTerm2 alert, or a Notification hook. That answers "did it stop?" for one agent you're actively watching. It doesn't answer "what should it do next?", and once you're running several agents a bell per tab is noise, not triage. The gap isn't the alert — it's that the alert can't carry a decision, and doesn't give you one queue for a fleet.

Route A — do it yourself

ntfy + a Claude Code hook

This genuinely works and it's free. ntfy is HTTP pub-sub with a phone app; a topic name is the address. Publish with one curl:

publish a push

curl -d "Claude Code needs your input" ntfy.sh/my-agents

Fire it from Claude Code's Notification hook so it triggers when the agent pauses for permission or stops:

~/.claude/settings.json

{ "hooks": { "Notification": [
  { "hooks": [ { "type":"command",
    "command":"curl -d \"agent needs you\" ntfy.sh/my-agents" } ] }
] } }

Subscribe to my-agents in the ntfy app. Done — you'll get a buzz. The limits are real, though: it's per-machine wiring you maintain, the topic name is the only "auth", and crucially the push can't carry a decision. It tells you to go look; you still open the terminal to actually answer.

Route B — Meltbox

The notification is the decision

In Meltbox, the way an agent talks to you is by pushing a brief — a self-contained page with clickable choices and a note box. It lands in your inbox and, on your phone, arrives as a push. You answer from the reader and your choice flows straight back to the waiting agent.

1 · connect the agent (once)

# App → Settings → Invite an agent → copy the one-time link, hand it to Claude Code.
# It redeems a scoped, revocable token — nothing to paste, no keys to copy.

2 · the agent says hello, then watches for you

mb brief push ./ship-plan.html --title "Deploy hotfix now, or wait for review?" --kind options
mb brief watch br_8f3k --since 0 --interval 3   # your answer streams back as events

3 · you get a push, and you answer it

[push] Your agent needs a decision — "Deploy hotfix now, or wait for review?"
# tap it → the brief opens → pick a choice → the agent proceeds. From the lock screen.

Add Meltbox to your home screen and allow notifications; push works on Mac and iPhone, including when the app is closed. Unlike a bare ping, the agent keeps a monitor on the brief, so the moment you tap a choice it does the next thing.

	ntfy hook	Meltbox brief
Gets a push	Yes	Yes
You can answer from the push	No — go to terminal	Yes — pick a choice
Answer feeds the agent	No	Yes — it's the next step
Works closed-app	depends on app	Yes (Mac + iPhone)
You maintain it	hook per machine	nothing on your side

Meltbox flow verified against docs.meltbox.ai/quickstart and Authoring briefs, July 2026.

FAQ

Common questions

How do I get notified when Claude Code needs permission?

Two routes. DIY: wire ntfy into Claude Code's Notification hook so it fires a push when the agent stops or asks. Meltbox: your agent pushes a brief, and with Meltbox installed to your home screen your phone gets a push you can answer from the lock screen.

Does the notification also let me respond?

With a plain ntfy or Slack ping, no — you go to the terminal to answer. With a Meltbox brief, the notification opens a clickable page; your choice flows back to the agent as its next step, so responding is the notification.

Does it work when the app is closed?

Yes — Meltbox is an installed PWA with real push on Mac and iPhone, including when the app is closed.

Do I need to install an SDK?

No. You install nothing beyond adding Meltbox to your home screen. The agent redeems a one-time invite link and pushes briefs; you triage them.

Turn "go check the terminal" into a push you can answer

Create a free workspace Watch an agent say hello

Or hand this to your own agent right now:

Read https://docs.meltbox.ai/quickstart and https://docs.meltbox.ai/invite-an-agent, then walk me through: creating a free Meltbox workspace at meltbox.ai, minting an invite link (Settings → Invite an agent), redeeming it with the mb CLI, and pushing me a first brief about what we're working on — then watch it for my reply.

More guides

Approve agent actions from your phone →Decide from the lock screen, not the terminal. Zero-inbox for agent output →Triage many agents like email. Meltbox vs. building it yourself →ntfy + Slack + .env + a board, honestly costed. Share one API key across agents →Use a key without any agent seeing it.

Sources

ntfy publish/subscribe model and app: ntfy.sh (verified Jul 2026).
Claude Code hooks reference: docs.claude.com/en/docs/claude-code/hooks.
Meltbox brief loop & triage keys: Quickstart, Authoring briefs (verified Jul 2026).