Setup backend in 5 mins

0:00/0:00

This video shows how to stand up a backend for an app in under five minutes with DYPAI. The example is a simple task manager—enough to see how everything fits together.

What you start with

In DYPAI you already have a project, but it can still be empty: no tables or endpoints yet.

Open the org "Conectar IA" (MCP tokens) page at /{org}/mcp/tokens, create a token, and copy the MCP config it gives you into your AI client. DYPAI works with Claude Code, Codex, Cursor, Windsurf, and Antigravity (and any MCP-compatible tool — the video uses Cursor). Once configured, DYPAI tools should show up for your assistant.

No IDE? Use Studio

You can also do all of this without an editor. Studio is the in-dashboard AI app builder and the default mode of every project — describe your app in a chat, watch the live preview, and publish. See Build an app in Studio.

Ask your AI agent to implement the full login flow for the app. The tutorial uses Composer 1 on purpose: it is fast and not the “smartest” model, to show that DYPAI stays versatile and quick to build with. Paste your prompt and run.

Your repo can be almost empty—for example Next.js (you could use another framework or even target mobile; Next is just a common choice). The agent will use DYPAI tools, read what it needs, and install missing packages for you.

When the first phase finishes, the agent will remind you to set the public URL of your DYPAI project: DYPAI → Connect, copy the project URL, put it in your environment variables (for example .env.local in Next.js), save, restart the dev server, and open the app.

You should see a login screen. Register a user: DYPAI sends a verification email. Confirm your account and you land inside the app—auth is live.

Phase 2 — App features

Go back to the agent and ask it to build the rest of the application (task features for this example).

While it works, watch DYPAI Web:

Database — tables the agent creates (maybe a single simple table at first).
API — endpoints appear as they are generated. They are not “plain REST only”: in DYPAI they are built as workflows tied to your database, with room for integrations. The engine already ships 40 workflow nodes, including ~24 third-party integrations like Stripe, Telegram, WhatsApp, Slack, Notion, Google Sheets, GitHub, and Shopify, plus a generic HTTP Request node for anything else.

The agent can validate and test endpoints as it goes; if something is wrong, it iterates and fixes errors.

Try it end-to-end

Create a test task, toggle it active, then open Database and confirm the row. In a few minutes you have authentication, first tables, and connected endpoints.

To go further, keep iterating with your AI agent and DYPAI—same pattern for storage, deeper settings, more complex workflows, payment gateways, and more. More videos are on the way.

What you'll learn

Wiring MCP and seeing DYPAI tools in the IDE
Letting the agent build auth and task flows from a minimal app
Setting .env with your DYPAI project URL and restarting the app
Watching tables and workflow-based endpoints appear in the dashboard

Next steps

Continue to What is DYPAI? for the big-picture story, then the rest of the series.

Connect MCP

No IDE? Use Studio

Phase 1 — Login with the AI

You should see a login screen. Register a user: DYPAI sends a verification email. Confirm your account and you land inside the app—auth is live.

Phase 2 — App features

Go back to the agent and ask it to build the rest of the application (task features for this example).

While it works, watch DYPAI Web:

Database — tables the agent creates (maybe a single simple table at first).

API — endpoints appear as they are generated. They are not “plain REST only”: in DYPAI they are built as workflows tied to your database, with room for integrations. The engine already ships 40 workflow nodes, including ~24 third-party integrations like Stripe, Telegram, WhatsApp, Slack, Notion, Google Sheets, GitHub, and Shopify, plus a generic HTTP Request node for anything else.

The agent can validate and test endpoints as it goes; if something is wrong, it iterates and fixes errors.

Try it end-to-end

Create a test task, toggle it active, then open Database and confirm the row. In a few minutes you have authentication, first tables, and connected endpoints.