Content
# MCP - Multi-Agent Collaborative Platform
[](https://github.com/LiamKujawski/MCP/actions/workflows/ci.yml)
[](https://github.com/LiamKujawski/MCP/actions/workflows/multi-agent-experiment.yml)
[](https://codecov.io/gh/LiamKujawski/MCP)
[](https://opensource.org/licenses/MIT)
## 🚀 Overview
MCP implements a **fully automated Research → Synthesis → Experiment → Deploy → Optimize loop** that transforms multi-model AI research into production deployments without human intervention.
### Key Features
- **🔄 Continuous Loop**: Automated pipeline from research to production
- **🧠 Multi-Model Synthesis**: Combines insights from O3, Claude-4-Sonnet, and Claude-4-Opus
- **🏗️ Full-Stack Generation**: FastAPI backend + Next.js frontend with ≥90% test coverage
- **📊 Real-Time Monitoring**: WebSocket-based UI for pipeline visualization
- **🐳 Multi-Arch Deployment**: ARM64 and AMD64 support with Traefik routing
- **🔒 Security First**: Semgrep/Bandit clean builds, OWASP compliance
## 🏛️ Architecture
### The Five Phases
```mermaid
graph LR
A[1. Research] --> B[2. Synthesis]
B --> C[3. Experiment]
C --> D[4. Deploy]
D --> E[5. Optimize]
E --> A
```
1. **Research Collection**: Normalized markdown with front-matter
2. **Synthesis**: AI-generated implementation prompts
3. **Experiment**: Parallel model execution and testing
4. **Deploy**: Automated containerization and deployment
5. **Optimize**: Performance monitoring and feedback
### System Components
- **Backend**: FastAPI with async Python, WebSocket support
- **Frontend**: Next.js 13+ with TypeScript, Tailwind CSS, Storybook
- **Testing**: Playwright E2E, Jest unit tests, pytest integration
- **Infrastructure**: Docker, Kubernetes, Traefik, GitHub Actions
- **Documentation**: C4 diagrams, ADRs, automated API docs
## 🚦 Quick Start
### Prerequisites
- Python 3.11+
- Node.js 20+
- Docker 24+
- Git
### Local Development
```bash
# Clone the repository
git clone https://github.com/LiamKujawski/MCP.git
cd MCP
# Backend setup
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
pip install -r requirements.txt
# Start backend
uvicorn src.main:app --reload
# Frontend setup (in new terminal)
cd ui
npm install
npm run dev
# Access the application
# Backend: http://localhost:8000
# Frontend: http://localhost:3000
```
### Running Tests
```bash
# Backend tests
pytest --cov=src --cov-report=html
# Frontend tests
cd ui
npm run test
npm run test:e2e
# Security scans
bandit -r src/
cd ui && npm audit
```
## 📁 Project Structure
```
MCP/
├── .github/workflows/ # CI/CD pipelines
│ ├── ci.yml # Main CI pipeline
│ └── multi-agent-experiment.yml # Research loop
├── chatgpt-agent-research/ # Agent research docs
├── codebase-generation-prompt-research/ # Prompt research
├── docs/ # Documentation
│ ├── architecture/ # C4 diagrams
│ └── LOOP.md # Pipeline documentation
├── src/ # Backend source code
├── ui/ # Frontend Next.js app
├── scripts/ # Automation scripts
├── docker-compose.prod.yml # Production config
└── synthesize-research-prompts/ # Generated prompts
```
## 🔬 Research Structure
Each research topic follows this structure:
```
research-type/
└── model-name/
├── 01_overview.md
├── 02_architecture.md
├── 03_prompt_design.md
├── 04_codebase_setup.md
└── 05_enhancements.md
```
With front-matter:
```yaml
---
topic: "topic-name"
model: "model-name"
stage: research
version: 1
---
```
## 🤖 Adding Research
1. Create markdown files following the naming convention
2. Include front-matter and DocOps footer
3. Push to main branch
4. The automation loop triggers automatically
## 📊 Monitoring
Access the monitoring dashboard at `http://localhost:3000/dashboard` to view:
- Pipeline execution status
- Model performance metrics
- Deployment health
- Research coverage
## 🚀 Deployment
### Production Deployment
```bash
# Build and deploy
docker-compose -f docker-compose.prod.yml up -d
# Scale services
docker-compose -f docker-compose.prod.yml scale web=3
```
### Environment Variables
Create `.env` file:
```env
DATABASE_URL=postgresql://user:pass@localhost/mcp
REDIS_URL=redis://localhost:6379
SECRET_KEY=your-secret-key
OPENAI_API_KEY=your-openai-key
ANTHROPIC_API_KEY=your-anthropic-key
```
## 📚 Documentation
- [Loop Documentation](docs/LOOP.md) - Complete pipeline guide
- [C4 Architecture](docs/architecture/) - System diagrams
- [API Documentation](http://localhost:8000/docs) - Interactive API docs
- [Storybook](http://localhost:6006) - Component library
## 🧪 Experimentation
Run experiments manually:
```bash
python scripts/run_experiment.py --model o3 --type full
```
Evaluate results:
```bash
python scripts/eval.py --output summary.json
```
## 🤝 Contributing
We follow the automated DocOps approach:
1. Add research following the structure
2. Let the system generate implementations
3. Review and iterate on results
4. Document learnings
## 📈 Performance
- **Pipeline Speed**: < 10 minutes from research to deployment
- **Test Coverage**: ≥ 90% across all components
- **Lighthouse Score**: ≥ 90 for UI performance
- **Security**: Zero HIGH/CRITICAL vulnerabilities
## 📄 License
This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.
## 🙏 Acknowledgments
- Research synthesis inspired by multi-model collaboration patterns
- Infrastructure patterns from cloud-native best practices
- Documentation approach based on Diátaxis framework
---
**Built with ❤️ by the MCP Team**
