How Retool's AI App Builder and AI Agents Work

Retool's AI app builder turns a written prompt into a full, production-ready internal app, and its AI agents let you build and deploy apps through an MCP server or CLI inside your governed Retool environment. Both paths assume you already have a Retool account and access to your production data sources; the output is meant to pass security review from day one rather than be rebuilt later.

What the AI app builder does

You describe the app you want in plain language, and Retool generates a complete app with enterprise security and governance built in. According to Retool, the result is "production-ready" from the start — authentication, access controls, and audit logging are already in place, so there is no separate hardening pass before shipping.

The builder is positioned for both technical and business teams: business users can move fast while IT keeps visibility and centralized governance.

Including data in a prompt with @

Inside the prompt, you use @ to reference and include data. This is how the generated app is wired to real sources rather than placeholder content. Because every Retool app connects directly to production data sources and access is governed by your existing permissions, the data you pull in through @ inherits those same permission rules — Retool states no extra configuration is required.

A practical example: if you need an operations dashboard for warehouse staff, you would prompt for the layout and use @ to attach the relevant production tables or APIs, and the app would respect each viewer's existing access level.

Building with AI agents via MCP server or CLI

Beyond the prompt-based builder, Retool supports agent-driven development:

• MCP server or CLI — you use either to build an app and then deploy it into your governed Retool environment.
• React code import — apps built on platforms like Lovable or Replit can be deployed by uploading a ZIP file or syncing with GitHub.

The key distinction is where the app lands: agent-built and imported apps are deployed within Retool's governed environment, so they inherit the same auth, access controls, and audit logging as apps built in the builder.

What "governed environment" means in practice

Retool frames governance as the reason AI-generated apps are usable at all. The platform's stated model:

Retool says more than 10,000 teams use it to generate production-ready AI applications, and cites customers such as Ramp (reported $8M and 20,000+ hours saved) and DoorDash (reported $6M and 36,000+ hours saved) — figures that come from Retool's own case stories and should be read as vendor-reported.

How to choose between the two paths

• Use the AI app builder when you can describe the app and its data needs in a prompt and want a finished app quickly, including for non-developers.
• Use MCP/CLI agents when you want to generate or iterate on an app programmatically and deploy it into the governed environment.
• Use code import when the app already exists on another platform (Lovable, Replit, or a GitHub repo) and you want to bring it under Retool's security and governance.

All three end in the same governed deployment, so the choice is about how you like to build, not about differing security outcomes.

Getting started and what to check

Retool offers a free way to explore the new app builder — the site's call to action is to "explore the new Retool app builder for free." Pricing details for paid plans are listed on Retool's pricing page; check that page directly for current tiers, since plan limits and feature gating are not specified in the material summarized here.

Before you build, confirm two things: that the data sources you plan to reference with @ are already connected in your Retool instance, and that your existing permission model covers the users who will open the app. Both determine whether the generated app is immediately usable or needs adjustment.