AI Agents · MCP

Build products your AI agents can actually use.

Ship a Model Context Protocol server with your app, so Claude and ChatGPT can call your tools — authenticated with your app's own accounts. One config, one deploy.

What you can build

Outcomes, not infrastructure

An agent-operable admin

Let an AI agent run your back office — read, create, update, publish — behind your real authentication and policies.

A customer-facing copilot

Embed an assistant that acts on your product's data and workflows, not one that only chats about them.

Internal MCP tools

Expose internal systems to your team's agents without standing up a separate service to maintain.

How it works on Maravilla

The APIs that power it

MCP Servers

Drop a file in mcp/ and it becomes an agent tool — auto-discovered, scoped, deployed with your app.

// maravilla.config.ts
mcp: {
  enabled: true,
  scopes: ['admin:manage'],
}

Auth & Connected Apps

Agents sign in with your app's own accounts; users see and revoke them from their profile.

const s = await platform.auth.login({
  email, password,
});

Authorization

Per-resource policies decide exactly what a tool — or an agent — is allowed to touch.

// policy on 'orders':
auth.user_id == node.owner || auth.is_admin

Ready to Simplify Your Workflow?

Start deploying in minutes. No credit card required.

Create Free Account