Mystic Writer : An Experiment on Agentic Development 🤖⚡

#ai #agents #fullstack #automaton

The Why ? 🤔

As a software developer exploring backend systems and automation, I wanted to test a bold idea :

Could I build a full-stack AI product - Backend, Authentication, Database, AI generation entirely through Agents, without writing a single line of code ?

The question became MysticWriter 🪶, an AI-powered story collaboration web application built in just two days using InsForge, Claude Haiku 4.5, and Figma Make.

My goal was to see what happens when developers stop coding line by line and start prompting their backends into existence.

What I built 🏗️🚧

MysticWriter lets users collaboratively write stories with AI, where user creates a story and the next line will be generated with AI and then along the way with back and forth communication between the humans and AI a complete story is created. It also generates a visual avatar for characters in the story as per need to have a visual effect of story.

All backend functionality such as authentication, database, story creation, bucket storage for generated images, and image generation was handled inside InsForge, an agentic-native BaaS that integrates seamlessly with AI through MCP (Model Context Protocol)

The frontend was designed in Figma Make, from prompts created the complete design and exported the design code to GitHub directly, just cloned the created Repository and then installed dependencies and started my frontend poof the design from figma is now in my local system.

The entire build from idea to working prototype happened over 48 Hours.

Here are some screenshots of the application 👇

Agentic Workflow 🪢🦾

Here's how the workflow actually unfolded :

I created an Insforge account and connected my MCP server from Insforge to Windsurf just with a simple command

 npx @insforge/install --client windsurf \
 --env API_KEY=your-api-key \
 --env API_BASE_URL=http://xxxxxxxx.insforge.app

Inside Windsurf, I used Claude Haiku 4.5 to issue structured prompts describing what I wanted InsForge to build things such as :
- "Create tables for users, stories and images. "
- "Add authentication and bucket storage for story images".
- "Setup AI collaboration for story and image generation".
Through the MCP connection, Claude communicated directly with InsForge tools to generate these backend resources, everything automatically just with AI-agents.
Whenever something went wrong (like field mismatch or auth misconfiguration), I'd prompt Claude again, and it would iterate or regenerate the setup.

Because Insforge is agent-native, everything from database creation to function deployment was prompt-driven! no SDKs, no CLI, no manual config.

Reflections After 48 Hours

This project was an eye-opener on how far agentic development has come.

Here’s what stood out:

✅ What Worked Really Well

Agent-native design: Insforge’s backend is natively accessible to AI agents through MCP! no hacks, no middle layers.
Instant visibility: The Insforge dashboard shows all created tables, functions, and AI credits, so I could track everything in real time.
Speed: From zero to a functioning app — with authentication, storage, and AI endpoints — in under two days.

What I missed ?

VS code as client : If Insforge has a VS code client along with Winsdsurf, Trae, Cursor, Claude Code, Cline it would be a nice to have.
BYOK : Bringing my own API Key for the AI models usage would be a nice to have feature, when I already have Keys for some models.

5. Tech Stack Overview

Tool / Platform	Purpose	Notes
Insforge	Agent-native BaaS	Handled auth, DB, AI calls, bucket storage, and function execution — all through prompts
MCP (Model Context Protocol)	Communication bridge	Enabled Claude to interface directly with Insforge from Windsurf
Windsurf	Agent workspace	Hosted the connected MCP server where I interacted with Claude
Claude Haiku 4.5	AI development assistant	Generated backend logic, fixed errors, and handled prompting
Figma Make	Frontend builder	Created the UI and connected it to Insforge APIs
MysticWriter	The final product	An AI storytelling app built entirely via agentic workflows

The Result

By the end of Day 2, MysticWriter was live and working:

Users could sign up, write stories, and generate illustrations.
AI endpoints handled the creative generation.
All data and media were securely stored through Insforge’s bucket and DB setup.

You can:

🌐 Try the Live Demo
💾 View the GitHub Repo

It’s a small MVP ,but a complete one and built entirely by prompting, not programming.

The Future of Agentic Development

Building MysticWriter in just two days fundamentally changed how I view development.

This wasn’t just “no-code"! It was agent-native.

With Insforge, MCP, and AI copilots like Claude, developers can focus purely on ideas and intent, while the agents handle the infrastructure and logic.