Build your first app
You have an idea. You want a real deployed app — auth, AI, database, file storage, a live URL — without stitching five services together. Here’s the whole path.
1. Run this in a terminal
The quickstart command walks you through everything you need:
- Checks for GitHub CLI (install it first if you don’t have it:
brew install ghon macOS,winget install GitHub.clion Windows, then rungh auth login). - Creates a free Recursiv account if you don’t already have one — you’ll paste an email and a 6-digit code from your inbox.
- Asks which AI client you’re using (Claude Code, Claude Desktop, Codex, or OpenCode) and installs the Recursiv MCP server into that client’s config.
About 3 minutes. When it finishes, fully quit and reopen whichever client you picked so the MCP loads.
2. Paste this into your AI client
Open your AI client in an empty folder. Paste this as your first message, replacing the bracketed idea with your own:
When the first tool-approval dialog appears, click Always allow recursiv — otherwise the client pauses every few seconds.
3. Wait, then open your URL
The build takes about 20 minutes the first time. When the agent prints the URL, it looks like https://<your-slug>.on.recursiv.io. Open it — that’s your app.
That URL is permanent. Every later deploy updates it in place.
After the URL
Keep chatting in the same window:
- “Add Google sign-in.”
- “Add a Stripe checkout.”
- “Add a notes field to the contact form.”
- “Connect my Gmail via
r.integrationsso the assistant can draft follow-ups.”
The agent edits, redeploys, and your URL updates.
Pricing
Free to start with $5 in credits, no credit card required. Builder plan is $49/mo + usage when you outgrow free.
If you get stuck
Ask the agent “what are you doing right now?” — it’ll narrate and resume. For tool errors, paste the error back and ask the agent to retry; the MCP tools are idempotent and most carry actionable hint fields the agent can read.