An agent can take an app all the way to a live address.
Most tools stop once the agent has written the code. Exepad keeps going: an agent creates the app, changes it, publishes it, and reads its logs — all from one key you hand over once.
Requires Pro and a funded balance · Published in minutes
What normally stops you
An agent can scaffold a project in seconds — then it stalls. Someone still has to give it a database, connect a sign-in, put it on a real address, and check that any of it works. That last mile is the part no agent owns, so a person never actually leaves the loop.
What changes for you
Four things that work differently once you build this way.
The whole lifecycle, as tools
Before
Your agent hands back a folder of code. Everything after that — a database, a sign-in, somewhere people can reach it — is a person at a terminal.
After
Tools cover the whole job: create an app from a description, change it, publish it, and read back its logs, errors and health. The agent finishes what it started.
Look before it edits
Before
An agent editing an app it can't see is guessing — it rewrites the page it was meant to extend, and you find out from the diff.
After
It reads the app's current setup back first, then makes a change that fits what's already there. Look, then edit.
The agent builds; you hold the wallet
Before
Handing an agent a payment method is why most people never hand one anything. The blast radius is your card, not your budget.
After
A key spends only from the balance you funded. It can't top that balance up, and it can't touch your payment method. You set the ceiling; the agent works under it.
Reach it however your agent talks
Before
Agent frameworks each assume one protocol, and half of what you run doesn't speak it.
After
MCP, A2A, or plain REST — the same lifecycle on all three, the same key, the same apps. Your agent uses whichever protocol it speaks; nothing is locked to one.
One app. Three ways to build it.
One app, three ways to build it: describe it yourself, call the API from your own program, or hand over the key and read the logs afterwards. The lanes meet at the same published address.
- 1Describe the site or tool you want, in a sentence or two.
- 2Read what comes back and ask for changes until it says what you mean.
- 3Publish it to a live web address and send people the link.
- 1Build the same app from your own program, not the chat.
- 2Create it and change it with calls, not by editing files.
- 3Publish it, read the logs, and roll it back if you need to.
- 1Hand an agent your key and let it build the app for you.
- 2It reads the app back, makes the changes you asked for, and publishes.
- 3Your balance and your spending limits stay yours alone.
One real, shareable address. No export step — just live.
What you build
Start from a category instead of a blank page.
The surfaces
One key, three ways to reach it. REST, MCP and A2A all run the same lifecycle against the same apps — create, change, publish, observe — so an agent uses whichever protocol it speaks.
REST
https://api.exepad.com/v1
The full lifecycle over plain JSON — create from a description, change, publish, and read logs. Builds stream over SSE; a Bearer key on every request.
MCP
https://mcp.exepad.com
The same full lifecycle as callable tools, for any agent that speaks the Model Context Protocol — discovered, not documented.
A2A
https://a2a.exepad.com
Agent-to-Agent messaging for any agent that speaks it. The same full lifecycle — create an app from a description, change it, publish it, observe it — driven by messages against the same apps.
But what about…
What does an agent need before it can build anything?+
An active Pro subscription and a funded wallet balance, or auto-reload enabled. Usage is billed per operation from that balance. Mint a key in your account, give it to the agent, and it can create, edit and publish apps until the balance runs out.
Does every app my agent publishes become an MCP server?+
No, and we would rather say so plainly than let you find out mid-integration. MCP is how an agent builds apps on Exepad — it is not something a published app turns around and exposes. A live app has no MCP endpoint to call.
Can an agent spend money I have not already put there?+
No. A key spends from your balance and cannot add to it or touch your payment method. That is the whole boundary: the agent decides what to build, you decide how much there is to build with, and those two decisions stay on opposite sides of the key.
Can an agent undo a change it got wrong?+
Not on its own. There is no restore tool in the agent surface, so an agent cannot put an app back to an earlier version. Version history and restore live in Studio, where a person does it. Plan for a human on that step, because there is one.
What if my agent does not speak MCP?+
Use whichever protocol it does speak. Plain REST for anything that can post JSON, or A2A for agent-to-agent — both run the same lifecycle on the same key: create from a description, change it, publish, and pull logs, errors and health.
How do I know what the agent actually did?+
Every edit streams its log while it runs, and you can pull logs, errors and health for an app afterwards over either surface. Webhooks push the same events to you. When you want to inspect the result yourself, the app is in Studio like any other.
Free covers simple web apps — three of them, one live at a time. The developer API is not part of it. An agent needs a key, and a key needs Pro.
Pro is $25/month and is what a key requires: unlimited full-stack apps, advanced AI models, and a balance the agent draws down per operation as it works.
See plansHand over the key. Get a running app.
One key, the whole lifecycle as tools, and an agent that finishes the job — from a sentence to a live address, without you in the loop.
Get an API key →