The backend your AI already knows how to use.
Project-438 plugs into Claude, Cursor, and other AI coding tools via MCP. Your AI sets up the database, wires up auth, configures storage — and writes the client code to talk to it. You don't touch a terminal.
// You, to your AI: "Build a task app with users, due dates, and file attachments." // Your AI, via P438 MCP: await projects_bootstrap({ name: 'task-app', migrations: [ { name: 'create_users', sql: '...' }, { name: 'create_tasks', sql: '...' }, ], }) // Project, schema, auth, storage — ready. // SDK code generated for your frontend.
AI builds the frontend in minutes. The backend still takes a week.
Vibe coding solved the UI. But every app needs a database, auth, file storage, an API. Your AI can write the SQL — but it can't run it anywhere, can't provision a project, can't tell you what tables exist. So you stop the flow, open a terminal, and wire it up by hand.
You prompt Claude or Cursor. It generates the UI in seconds. Then you spend the rest of the day setting up Postgres, writing migrations, configuring auth providers, picking a storage solution, gluing it all together. By the time your "5-minute app" is real, it's the next morning.
The backend is the bottleneck.
Your AI tool already knows how to operate Project-438 — through MCP, the same protocol it uses for filesystem access and search. It creates the project, applies your schema, configures auth and storage, and hands you working SDK code. All you do is describe the app.
Stay in the flow. Ship in one sitting.
Three steps to a working backend.
Connect P438 to your AI tool
Drop a few lines into your Claude Desktop or Cursor config. The Project-438 MCP server gives your AI a real backend it can operate — create projects, design schemas, run migrations, inspect data.
Describe what you want to build
Tell your AI the app you have in mind. It picks the schema, applies the migrations, sets up auth and storage — using P438 tools, not by writing infrastructure code.
Your AI writes the client code
Project-438 generates working TypeScript snippets for the tables and operations your app needs. Your AI pastes them into your frontend. Run it. It works.
Get early access.
Looking for vibe coders, indie devs, and small teams who want their AI to handle the backend too. Alpha access goes out in batches.