Catwalk Live

A Vercel-like platform for deploying Remote MCP servers to Fly.io, built entirely through AI orchestration.

🤖 This project was built using a multi-stage AI development pipeline (Claude Code, Cursor, Gemini Code Assist) without manually writing code. It demonstrates strategic prompt engineering, multi-agent coordination, and quality control of AI-generated systems.

---

What It Does

Catwalk Live makes deploying Model Context Protocol (MCP) servers as simple as deploying to Vercel:

1. Paste a GitHub repo URL → AI analyzes the MCP server automatically
2. Enter credentials → Securely encrypted with Fernet
3. Click deploy → Spins up isolated Fly.io container
4. Get endpoint → Connect Claude Desktop immediately

Live Demo: Video Walkthrough (coming soon)

---

Features

• 🔍 Auto-Analysis: Paste any GitHub MCP server repo → AI extracts config (tools, resources, prompts, env vars)
• 🔐 Secure Credentials: Fernet encryption for API keys/tokens stored in PostgreSQL
• 🚀 One-Click Deploy: Automated deployment to Fly.io with isolated containers
• 🌐 Streamable HTTP: Implements MCP 2025-06-18 spec (latest standard)
• 📊 Real-Time Status: Track deployment health and logs
• ✅ Package Validation: Checks npm/PyPI registries before deployment
• 🛡️ Input Validation: Prevents SQL injection, XSS, command injection

---

🤖 Built with AI Orchestration

This project is a case study in AI-assisted development. Here's the workflow:

Development Pipeline

Prompt Refinement (Plain English → Detailed Spec)
         ↓
Multi-AI Planning (Claude + GPT-4 + Gemini)
         ↓
Implementation (Claude Code + Cursor)
         ↓
Quality Gates (CodeRabbit, Qodo, Gemini Code Assist, Greptile)
         ↓
Iteration (Feedback → AI → Fixes)
         ↓
Production Deployment (Fly.io)

Key insight: AI needs structure to build production systems. Context files (AGENTS.md, context/*.md) act as "external memory" that persists across AI sessions.

Meta-Documentation (The Real Value)

This repo contains not just code, but a reproducible methodology for AI-orchestrated development:

• AGENTS.md - AI agent specifications & interaction protocols
• context/ - Knowledge base for guiding AI development
• CLAUDE.md - Lessons learned, deployment pitfalls, debugging patterns
• AI_ORCHESTRATION.md - Full methodology case study (read this!)

See AI_ORCHESTRATION.md for the complete story - how to build production systems with AI, what works, what doesn't, and lessons learned.

---

Tech Stack

Frontend: Next.js 15 (App Router), React 19, TailwindCSS 4, TypeScript 5+ Backend: FastAPI (Python 3.12), SQLAlchemy (async), PostgreSQL 15+ Infrastructure: Fly.io (Machines API), Docker MCP Transport: Streamable HTTP (2025-06-18 spec) Security: Fernet encryption, Pydantic validation, environment isolation

---

Quick Start

See SETUP.md for detailed instructions.

Backend

cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env  # Fill in API keys
alembic upgrade head
uvicorn app.main:app --reload

Backend runs at: http://localhost:8000

Frontend

cd frontend
bun install
cp .env.local.example .env.local
bun run dev

Frontend runs at: http://localhost:3000

---

Deployment

See DEPLOYMENT.md for production deployment to Fly.io.

TLDR:

# Create Fly apps
fly apps create <your-backend-app>
fly postgres create --name <your-database-app>
fly postgres attach <your-database-app> --app <your-backend-app>

# Set secrets
fly secrets set ENCRYPTION_KEY="..." OPENROUTER_API_KEY="..." --app <your-backend-app>

# Deploy
cd backend && fly deploy --app <your-backend-app>

Current deployment (as of December 2025): Backend: https://<your-backend-app>.fly.dev (replace with your deployment)

---

How It Works

1. User Analyzes GitHub Repo

User → https://github.com/user/mcp-server
     ↓
Claude analyzes package.json or pyproject.toml
     ↓
Extracts: package name, env vars, tools, resources

2. Dynamic Form Generation

Based on extracted env vars, frontend generates a credential input form:

{
  "env_API_KEY": { type: "password", required: true },
  "env_BASE_URL": { type: "url", required: false }
}

3. Secure Storage

Credentials encrypted with Fernet before storage:

cipher = Fernet(settings.ENCRYPTION_KEY)
encrypted = cipher.encrypt(json.dumps(credentials).encode())
# Stored in PostgreSQL, decrypted only at deployment time

4. Deployment to Fly.io

Creates isolated Fly Machine running:

• mcp-proxy (Streamable HTTP adapter)
• User's MCP server package (npm or PyPI)
• Injected credentials as environment variables

5. Claude Connects

Stable endpoint: https://<your-backend-app>.fly.dev/api/mcp/{deployment_id}

Claude Desktop → Backend → Fly Machine → MCP Server → Tool Execution

---

Architecture

┌─────────────────┐
│  Claude Desktop │
│   (MCP Client)  │
└────────┬────────┘
         │ Streamable HTTP
         │ (2025-06-18)
         ↓
┌─────────────────┐
│   Catwalk Live  │
│    (Backend)    │
│  FastAPI + PG   │
└────────┬────────┘
         │ Fly Private Network
         ↓
┌─────────────────┐
│  MCP Machine    │
│  mcp-proxy +    │
│  MCP Server     │
└─────────────────┘

See context/ARCHITECTURE.md for detailed system design.

---

Contributing

We welcome contributions - especially AI-assisted ones!

See CONTRIBUTING.md for:

• How to contribute using AI tools (Claude Code, Cursor, ChatGPT, etc.)
• Quality standards and automated review process
• Expectations for contributors

TL;DR:

1. Fork the repo
2. Use AI assistants with AGENTS.md as context
3. Submit PR (automated agents will review)
4. Iterate with AI based on feedback
5. Merge!

---

About This Project

I didn't write this code manually—I orchestrated AI systems to build it.

This demonstrates:

• ✅ Strategic prompt engineering for complex systems
• ✅ Multi-agent coordination across backend, frontend, infrastructure
• ✅ Quality control and validation of AI-generated code
• ✅ Shipping production-ready AI-assisted projects

I'm not a traditional backend engineer—I'm an AI Orchestrator and Technical Product Builder. I can't write FastAPI from scratch, but I can architect systems, validate AI outputs, catch integration bugs, and ship working products.

If you're hiring for:

• AI Engineering Manager roles
• Technical Product Management (AI tools)
• AI-Assisted Development positions
• Developer Experience (AI tools)

Let's talk! Email: jordanlive121@gmail.com

---

Project Status

Current Phase: Phase 1 Complete (Validation) + Phase 6 Working (Streamable HTTP)

What Works:

• ✅ Full backend deployed on Fly.io
• ✅ GitHub repo analysis with AI
• ✅ Package validation (npm + PyPI)
• ✅ Credential validation and encryption
• ✅ Deployment to Fly Machines
• ✅ End-to-end MCP tool calling

What's Next:

• Health monitoring loop
• Rich deployment progress reporting
• Frontend deployment (currently local only)

See context/CURRENT_STATUS.md for detailed status.

---

Resources

Documentation:

• SETUP.md - Local development guide
• DEPLOYMENT.md - Production deployment to Fly.io
• CONTRIBUTING.md - How to contribute (with AI!)
• SECURITY.md - Security policy and best practices
• AI_ORCHESTRATION.md - Full AI methodology case study

MCP Resources:

• Model Context Protocol Spec
• MCP Servers Repository
• Fly.io Machines API

---

License

MIT License - see LICENSE for details.

Copyright (c) 2024-2025 Jordan Hindo

---

Acknowledgments

Built with:

• Claude Code (Anthropic) - Primary AI coding assistant
• Cursor - Refactoring and iterative development
• Google Gemini - Planning and cross-validation
• CodeRabbit, Qodo, Gemini Code Assist, Greptile - Automated PR review

Inspired by:

• Vercel's developer experience
• The MCP ecosystem
• The future of AI-assisted development

---

⭐ If this project helped you understand AI orchestration, please star the repo!

🚀 Ready to build your own AI-orchestrated project? Read AI_ORCHESTRATION.md and start experimenting!