Connect your agent

Setup takes a minute, once. Then you ask for an app in your own words.

Apphaven is where the apps your AI agent builds live. Your agent is the one you already talk to, such as Claude, ChatGPT or Gemini. You connect it to your Apphaven account, and from then on it builds, tests and looks after apps for you, while you run them from your console.

Before you start

You need an Apphaven account: sign up or sign in. Building and testing are free, so there's nothing to set up for billing until you publish a live app.

Your agent will ask for Apphaven's connector address. It's the same for everyone:

https://apphaven.com/mcp

Your console's Connect your agent page has the same address with a copy button, plus a first request to try.

Add it in Claude

1. In Claude, open Settings, then Connectors.
2. Choose Add custom connector.
3. Name it Apphaven and paste the connector address.
4. Claude opens an Apphaven page. Sign in if asked, then choose Allow.
5. Start a new chat and ask for an app.

Once Apphaven is listed in Claude's connector directory, you can add it from there instead of pasting the address. Either way works the same afterwards.

Add it in ChatGPT

1. In ChatGPT, open Settings, then Apps & connectors. If you don't see a way to add your own, turn on developer mode under Advanced.
2. Add a new connector (it may be called an MCP server) and paste the connector address.
3. When Apphaven opens, sign in if asked and choose Allow.
4. In a new chat, turn on the Apphaven connector and ask for an app.

Add it in Gemini

1. Go to gemini.google.com on a computer and open Settings, then Connected apps. If you don't see it, open Personal Intelligence first.
2. Under Custom apps, choose Add a custom app and paste the connector address.
3. Choose Next. When Apphaven opens, sign in if asked and choose Allow.
4. Ask for an app. Custom apps connected on the computer work in the Gemini mobile app too.

Google currently offers custom apps to personal Google accounts in the US, in English, with Keep Activity turned on.

Claude Code, Cursor and other tools

Any tool that takes an MCP server can use the connector address. Most open Apphaven in your browser so you can sign in and choose Allow. For Claude Code, run this in your terminal, then type /mcp to sign in:

claude mcp add --transport http Apphaven https://apphaven.com/mcp

If a tool can't sign in through your browser, create an access token on your console's Connect page and give the tool the header Authorization: Bearer <your token>. A token works like a password for your account, so keep it private, and revoke it from the same page when you stop using the tool.

The approval screen

The first time an agent connects, Apphaven shows a page headed Connect your agent to Apphaven? It names the agent asking, the account it will act as, and where the access is being sent. Choosing Allow lets that agent build and run apps as you, until you disconnect it from the agent's own settings. Choosing Cancel sends it back with nothing.

Tools that run on your own computer, like Claude Code, get one more line: a warning that the access is being sent to a program on this computer. That's normal for those tools. Continue only if you just started connecting from that program.

What your agent can do

Everything your agent does goes through a fixed set of actions, and each one is marked as reading, writing or changing something live, so your agent (and Claude's permission prompts) know which ones to be careful with.

• Look at thingsRead its own guide, see your apps and their status, read an app's files, look up its data, list the people signed up to it, read its logs and the mail it sent, check a configuration, look at a page like a person would.
• Build and testCreate an app, write and edit its files, deploy the test version, try the app's own API as a signed-in person or as a visitor, and send the Apphaven team a note about what helped or got in the way.
• Change the live app, when you askPush the test version's changes to the live app, restore an earlier version, add or change records, and manage the people in the app. If a change would delete live data, the agent stops and you approve it in the console instead.

What stays with you

Some things only happen in your console, never through an agent: publishing an app for the first time and choosing a plan, putting an app on your own domain (the agent then walks you through the one record to add), deleting an app or your account, billing, and access tokens. Your agent can read your apps' data and change it when you ask; it can't reach your other apps' people or anyone else's apps.

Disconnecting

Remove the Apphaven connector in your agent's settings, and that agent loses access straight away. Access tokens are revoked from your console's Connect page. Your apps keep running either way.

For your agent: the guide it reads

Your agent reads Apphaven's guide itself, through the connector, so you never need to explain how Apphaven works. The same text is published here, for the curious and for anyone helping you:

• Overview: the workflow, files, sign-in, email, installable apps, your own domain.
• Configuration: data models, access rules, operations, sign-in settings.
• Client SDK and REST API: how an app's pages talk to its data.
• Files people send the app: photos and documents from the app's users.
• Example apps: complete apps, start to finish.

Help

Write to support@apphaven.com. If your agent gets stuck, the best first step is to ask it to read its guide again and tell you what it found.