- SPECIMEN
Pushary
- FILED AS
Lock screen approvals and push notifications for AI coding agents via MCP
- INTAKE DATE
- 2026-08-04
- CLASS
- Coding
- METHOD
- 1 page scraped, 3 search passes, 13 community sources
Koda Score = weighted blend: capability 35, ease 20, value 25, momentum 20.
Your agent stalls on a yes-or-no question while you are making coffee; Pushary moves that question to your lock screen and gets the run moving again.
Captured: mcpservers.org/servers/pushary/pushary-skill on 2026-08-04.
This is for people running two or more coding agents who keep losing minutes to approval prompts nobody is at the desk to answer. If that is you, the $9.99/mo is cheap against the dead time, and the MCP-plus-hooks install is genuinely a one-command affair with Claude Code. If you run a single agent on Claude Code with Claude Max, Anthropic's own Remote Control already covers that case for free and Pushary is redundant.
What it actually does.
Lock screen answer buttons
Agent questions arrive as notifications you can answer without opening into a terminal session.
Three question types
Agents pick yes/no confirmations, multiple choice, or free text depending on what they actually need from you.
Claude Code permission hooks
Routes native tool approval prompts out of the terminal and into push notifications.
Rich context notifications
Agents can attach file changes, error details and suggested next steps to a detail page behind the notification.
Agent identification
Each notification names the agent asking, so a fleet of parallel runs does not become a guessing game.
Cross-agent fleet coverage
One inbox spanning Claude Code, Codex, Cursor, Windsurf, Hermes, Lovable and any MCP client.
Audit trail
Every question and answer is logged, which is the part that matters when someone asks who approved what.
One-command plugin install
The repo ships as a Claude Code plugin that installs MCP tools, hooks and the skill together.
How you would actually use it.
Concrete setups pulled from the research, not feature-list hand-waving.
The coffee-run refactor
Solo builder running one long agent task
Install the Claude Code plugin, set PUSHARY_API_KEY, and tell the agent to keep going and ping your phone if it needs anything.
A 20-minute refactor no longer sits idle waiting for you to walk back to the desk.
Fleet triage from one inbox
Developer running Codex, Cursor and Claude Code in parallel
Point each agent at the hosted MCP endpoint with the same API key so every question lands in one notification stream with the agent name attached.
You stop tab-hopping between three tool UIs to find out which run is blocked.
Gated writes with a paper trail
Small team lead reviewing agent changes
Enable permission hooks so tool approvals route to phones, keep the safe-task list to read-only operations, and lean on the logged question-and-answer history.
Approvals happen in minutes instead of hours, and you can reconstruct who greenlit each action.
Done and failed alerts
Anyone kicking off overnight or background runs
Have the agent fire task-done and task-failed notifications instead of you polling the terminal.
Failures surface immediately rather than at the next manual check-in.
Pricing, straight.
7-day card-first trial, MCP support, cross-agent fleet, lock screen answer buttons
Ten dollars a month is fair if you genuinely run multiple agents, but it is a hard sell as a notification relay for a single Claude Max setup that Anthropic already covers for free.
Pricing as captured on 2026-08-04. Check the live site before you commit.
What people online are saying.
Real reactions surfaced during research. Paraphrased faithfully, linked to source.
Reaction is thin and mostly launch-adjacent: Product Hunt supporters love the idea of a phone yes button, and the sharpest public critique is that a lock screen strips away the session context you had 40 minutes ago.
Supporters like that it collapses an agent pause into a simple yes button on your phone, so the work keeps moving while you are away from the desk.
The launch thread frames task-done, task-failed and task-needs-decision alerts as the killer use case for keeping coding agents moving.
At the terminal the whole session is scrolled above you; on a lock screen you get one diff and none of the intent you set 40 minutes ago and have half-forgotten.
The field.
Anthropic's own remote approval path for Claude Code, covered by a Claude Max subscription.
PICK IT WHENYou run exactly one Claude Code setup on Claude Max and do not want another subscription.
Generic notification channels wired to agent events, which Pushary also supports as fallback delivery.
PICK IT WHENApprovals should live where your team already discusses them and lock screen speed is not the point.
Each agent's native in-tool permission prompts.
PICK IT WHENYou work from one machine at one desk and do not need cross-agent centralization.
The honest part.
- Requires an API key and a card for the trial; there is no free tier.
- Notifications only cover supported task types, so coverage depends on how your agent is instrumented.
- Installing both the Claude Code plugin and the npx agent-hooks setup runs every hook twice; you have to clean one path first.
- A lock screen prompt shows one decision without the surrounding session, which is exactly the context critics say gets lost.
- Independent reviews are close to nonexistent; almost all public material is launch-thread commentary.
Getting started.
Time to first value: about 10 minutes
Sign up at pushary.com, grab your API key and set PUSHARY_API_KEY in your environment.
Run /plugin marketplace add Pushary/pushary-skill then /plugin install pushary in Claude Code, or add the hosted MCP endpoint with a Bearer token to settings.json.
Enable notifications from the Pushary dashboard on your phone, then trigger a throwaway approval to confirm the round trip.
Quick answers.
Is Pushary actually free?
No. The site lists a single Monthly plan at $9.99/mo after a 7-day card-first trial, so plan on the subscription.
Do I need a Claude Max subscription?
No, Pushary works without one and explicitly targets setups Anthropic Remote Control does not cover: Codex, Cursor, Claude Code without Max, and multi-agent fleets. If you do have Claude Max and only use Claude Code, Remote Control already handles that one case for free.
Can it auto-approve safe tasks so I am not tapping all day?
The permission hooks route Claude Code tool approvals through notifications and the product describes per-tool policy gating, so you can decide which classes of action need a human. Keep that list conservative at first, since the research surfaces no independent testing of how the policy layer behaves under pressure.
What is the main risk of approving from a phone?
Context loss. The sharpest public critique, from the Product Hunt discussion, is that the terminal shows you the whole scrolled session while a lock screen gives you one diff and none of the intent you set earlier in the run. Rich context notifications with file changes and error details soften this, but they do not replace reading the session.
Wire it to one agent, keep the auto-approve list boring for a week, and only then let it near anything that touches production.
Field research: 1 pages scraped · 3 search passes · 13 community sources. Reviewed by the Koda desk on 2026-08-04.