How it works
You talk to the AI. The agent handles the rest.
Builder Agent is an MCP that connects to your AI tool — Lovable, Claude Code, Codex, Replit or AI editors — and provides tools that make the build follow an order: ask, measure the impact, build in parts, test with proof and document.
The path
From the first request to a live app.
01
Connect the MCP
to your AI.
Paste the Builder Agent address into your tool — Lovable, Claude Code, Codex, Replit or your AI editor — and sign in with your account. From then on, the tools show up for the AI.
The AI you already use, now with a method behind every answer.
02
Discovery
before code.
The AI interviews you: vision, type of app, who uses it and which screens exist. Nothing starts on an assumption — and the result becomes a document inside the project itself.
Your app's document written in the repository: vision, users and interfaces.
03
Implementation
guide.
Screens and features get queued: what comes first, what depends on what, the build order and the order of use. No loops, no orphan features.
A checklist roadmap: the order to build in and the order users will follow.
04
One feature
at a time, with a gate.
Every feature takes the same path: locate, measure the impact, approved plan, validation and close. No proof, no close.
A feature that actually works — tested, not “almost done.”
05
Audits
and memory.
Architecture, data, security, interface, features and production get audited with proof. And everything that's done is written into the project — tomorrow's memory.
An app that grows without falling apart: feature 10 doesn't break feature 3.
The main tools
Five paths.
Seventeen tools.
Each tool has its own input, output and lock — the AI can't skip a step. They're organized into five paths, from starting the project to fixing bugs.
Start the project
App discovery
The MCP guides your building AI through a complete interview tailored to your app.
Implementation guide
Sorts out what comes first and what depends on what, before building starts.
New interface
Creates the structure of a new screen following the project's patterns.
Build a feature
Feature path
Guides your AI to build one feature at a time, with validation and proof.
Closing gate
Only moves to the next step once the current one is truly done.
Audit
Full audit
Reviews the whole project and flags what needs attention before you scale.
Architecture and data
Checks the app's structure and data to keep everything consistent.
Security and production
Checks access, personal data and readiness to go live.
Interface and features
Reviews screens and features one by one to keep things consistent.
Adopt an existing project
Adopt the app
Documents the existing app so the tools understand the current project.
Adopt interfaces and features
Maps the screens and features already built so you can keep going the right way.
Adopt integrations
Catalogs the app's external services and connection points.
Fix a bug
Fix with proof
Finds the bug, fixes the minimum needed and requires proof that it passed.
The same seventeen tools come in dedicated versions for Lovable, Claude Code, Codex and Replit — each one speaking that tool's language. Your subscription sets which platform is included and which others you unlock.
Get started