Build apps from your AI assistant
Describe the app you need to Claude, ChatGPT or Gemini: your assistant writes it, checks it by actually compiling it, and hands it back as a link. You open it, and the app is in your Reactive library.
The same address for every assistant. It needs no account and no password: where the client asks for credentials or OAuth, leave it empty.
Connect your assistant
Claude
Free plan is enough- Settings → Connectors
- Add custom connector
- Paste the address
Works on every plan, Free included.
ChatGPT
Paid plan required- Settings → Apps → Advanced settings
- Turn on Developer mode
- Add the connector and paste the address
Plus, Pro, Business, Enterprise or Edu. Custom connectors do not exist on the Free plan.
Gemini
Paid plan required- gemini.google.com → Settings
- Connected apps
- Paste the MCP server address
Once connected on the web it also works from the mobile app.
Claude Code
Free plan is enough- claude mcp add --transport http reactive https://mcp.reactivenet.ai/mcp
One command, from the terminal.
On managed accounts an administrator must enable it first.
What the assistant can do once connected
- Read the syntax guide — The full Reactive language reference, section by section.
- Browse the directives — All 145 directives with their form and attributes.
- Start from a ready block — Twelve tested building blocks: CRUD list, registry, dashboard, kanban, bookings, stock…
- Validate the app — Compiles the document through the very pipeline the app uses and reports every problem with its line number.
- Hand you the link — Packs the finished app into a link: you open it, the app lands in your library.
What the server sees
Only the document being written. It holds no data, keeps no sessions and has nothing to log in to: your app data lives in your browser, exactly as when you build an app inside Reactive. The assistant, of course, sees what you type — that is between you and the assistant you chose.
Why it goes through a check
Reactive's language is made of directives with a precise syntax, and a model that gets one wrong would only find out when you open a broken app. The connector closes that gap: before handing you anything, the assistant compiles the document through the very pipeline the app uses and gets the errors back with line numbers. If it does not compile, it does not even get the link — it has to fix and retry.
It is the same check that runs in our continuous integration on every change to the language: when we add a directive, the connector knows about it the same day.
Rather not configure anything?
Reactive already ships an assistant that writes apps, and it runs on your own device. No account, nothing to connect.
Build your app