Agent asks
One typed MCP call creates a durable decision request.
DecisionPing gives AI agents one reliable MCP connection for human approvals, choices and text answers — delivered to the channels people already use.
Version 2.4.1 passed all checks and is ready to ship.
One typed MCP call creates a durable decision request.
The request reaches Telegram, WWChat and connected endpoints.
A structured, explicit answer returns to the waiting workflow.
DecisionPing began as a tool its author built for himself: kick off a long agent run, leave the desk, and trust that the agent will reach you when it matters.
Start a refactor, a migration, a test suite — and go to the gym, the store, the couch. If the agent hits a question it can't answer alone, it pings the app on your phone. You answer; it continues. No question? You just get a “done” report.
The laptop in your backpack runs nothing. Move long tasks to your own server with agentd: the chat and the agent live there, and your machine is free to sleep.
Cloud chats give you a managed workspace in seconds — Python, Go and the Android toolchain preinstalled. No rented VPS, no setup, no maintenance. Bring your agent sign-in and start typing.
The internet is full of “check out my SaaS — https://localhost:3000”. From a DecisionPing chat your agent builds and deploys your project with your own tools — so you share a real URL people can actually open.
Beyond approvals, DecisionPing runs full AI coding agents you can talk to — Codex, Claude and Grok — in a managed cloud workspace or on your own server.
Spin up a managed container in seconds, with Python, Go and the Android toolchain preinstalled. No server and no setup — just bring your own agent sign-in.
Prefer your own machine? Connect a Linux server, Mac or Windows PC with agentd and choose exactly how much access the agent gets.
Connect GitHub once and let chats clone and push to the repositories you choose.
DecisionPing started as a single notification: an agent working on my machine could ask me a question through my phone, instead of silently guessing or waiting for me at the keyboard. Then I kept leaving the office with my laptop in a backpack — and everything stopped. So the agents moved to a server. Then to a managed cloud, so nobody has to configure a server at all. The name stayed; the meaning grew. Today I run my own development through DecisionPing every day — every feature on this page exists because I needed it.
Read the devlog →Say: “Connect DecisionPing using decisionping.com/connect.” The agent-readable guide explains exactly how to configure the service safely.
Human-in-the-loop infrastructure for AI agents: a remote MCP server your agent calls when it needs an explicit human decision — an approval, a choice between options, or a typed answer — plus notifications when it just wants to report progress.
Requests are delivered to the DecisionPing app (iOS, Android, macOS, Windows), and to Telegram or WWChat if you connect them. You answer wherever is closest; the agent receives one structured response.
The request is durable: the agent waits (or polls) until you respond or the request expires — 30 minutes by default, configurable from a minute to a week. It never invents your answer.
Any MCP client: one small config block connects Claude Code, Codex, Cursor and others. Cloud chats additionally run full coding agents — Codex, Claude and Gemini CLI — for you.
No. Cloud chats work out of the box. If you prefer your own hardware, connect a Linux server, Mac or Windows PC with agentd and choose how much access the agent gets.
There is no subscription. Signing up is free; you top up a prepaid balance by card, and usage draws from it.