Build a Full-Stack App with a Single Sentence: The Complete InsForge Guide

“Tell an AI agent, ‘Make a to-do list with login,’ and watch the backend, database, and file storage appear automatically.”
This walk-through will show you—step by step—how to turn that wish into reality.

Table of Contents

1. What is InsForge, exactly?
2. What can it do for you?
3. Local installation in three terminal commands
4. Plug any AI agent (Claude, GPT-4o, etc.) into InsForge
5. From prompt to production: three real projects you can copy-paste
6. A five-minute tour of the architecture
7. Frequently asked questions (FAQ)
8. Where to learn more and get human help

1. What is InsForge, exactly?

InsForge calls itself “the Agent-Native Supabase Alternative.”
In plain English: it rebuilds the most common backend services—authentication, database, file storage, serverless functions, and one-click deployment—so that AI agents can manage them for you.

Instead of writing SQL, REST endpoints, or deployment scripts, you describe what you want in natural language. The agent then calls InsForge’s pre-built tools and returns working URLs you can drop straight into a front-end.

2. What can it do for you?

3. Local installation in three terminal commands

Prerequisites

• Docker Desktop (any OS)
• Node.js 18+ (only used by the CLI helpers inside the containers)

Step 1: Clone the repository

git clone https://github.com/insforge/insforge.git
cd insforge

Step 2: Copy the environment template

cp .env.docker.example .env

Leave the defaults for now; you can edit ports and secrets later.

Step 3: Start the stack

docker compose up

When the logs show

insforge-db | ready to accept connections

open http://localhost:7131 in your browser.

4. Plug any AI agent into InsForge

4.1 Turn on MCP inside InsForge

After you log in, the sidebar shows a “Get Started” wizard. Click it and look for the MCP section. Copy the URL that looks like

mcp://localhost:7131/mcp

4.2 Add the MCP server to your agent

Below are two popular examples; any client that speaks MCP works the same way.

Claude Desktop

1. Settings → MCP Servers → Add
2. Name: insforge
3. URL: paste the MCP link
4. Restart Claude

GPT-4 via custom backend

If you run GPT-4 through your own code, POST the MCP descriptor JSON to https://api.openai.com/v1/tools (or use the official SDK helper).

4.3 Test the connection

Send this prompt to the agent:

InsForge is my backend platform. What is my current backend structure?

A healthy response lists the available tools: createTable, insertRow, uploadFile, queryRows, etc.

5. From prompt to production: three real projects

Copy the exact prompts into your agent after you complete the MCP setup.

Project 1: To-do list with user accounts

Prompt

Build a to-do app with user authentication.

The agent will:

1. Create users table (id, email, password_hash, created_at)
2. Create todos table (id, user_id, title, done, created_at)
3. Generate REST endpoints:

   • POST /auth/register
   • POST /auth/login
   • GET /todos (auth-guarded)
   • POST /todos
4. Return a ready-to-paste front-end snippet for Lovable, Bolt, or plain React.

Project 2: Instagram clone with image upload

Prompt

Create an Instagram clone that allows image uploads.

The agent will:

1. Create users, posts, and likes tables
2. Call the UploadFile tool to store images in InsForge Storage
3. Return CDN URLs you can embed in <img> tags
4. Provide sample queries:

   • Fetch feed for user
   • Like/unlike a post

Project 3: Mini e-commerce MVP

Prompt

Build a small online shop with products, cart, and checkout.

The agent will:

1. Create products, orders, order_items, and customers tables
2. Add an UploadFile step for product photos
3. Generate checkout webhook placeholders (ready for Stripe or PayPal integration later)

6. A five-minute tour of the architecture

The diagram below shows how all the parts fit together.

7. Frequently asked questions (FAQ)

Installation

Q1. Docker fails to start with “port already in use.”
A. Edit .env, change 7131 or 5432 to free ports, then docker compose up again.

Q2. Can I run this on Windows?
A. Yes. Install Docker Desktop for Windows and follow the same commands.

Q3. Do I need a separate Postgres installation?
A. No. The compose file already includes an official Postgres container.

Usage

Q4. Could the AI accidentally delete my data?
A. The MCP tools run under a sandboxed role scoped to your current project. For production, create a read-only user and point the agent to it.

Q5. How do I add an index?
A. Tell the agent:

Add an index on todos.user_id

It will run CREATE INDEX idx_todos_user_id ON todos(user_id) and confirm success.

Q6. Is there a file-size limit?
A. 100 MB per file at the moment. The limit can be raised by editing an environment variable once the feature is publicly released.

Q7. How do I enable phone-number login?
A. Dashboard → Authentication → Providers → enable “Phone”. The agent will immediately see the new provider in the schema.

Q8. Does it support non-English text?
A. The database uses UTF-8. The AI agent supports any language its underlying model supports.

Q9. How do I back up the database?
A. Run:

docker exec -it insforge-db pg_dump -U postgres insforge > backup.sql

Automated scripts are on the roadmap.

Roadmap

Q10. When will serverless functions arrive?
A. Target is Q4 2024. Supported runtimes will be Python, Node.js, and Go.

Q11. Will I be able to deploy to Vercel or Netlify?
A. Yes. The upcoming “Site Deployment” feature will include one-click buttons for major platforms.

Q12. Is there a paid tier?
A. The local version is free forever. A cloud-hosted, pay-as-you-go version is planned; pricing will be announced later.