AI Brand Manager with MCP and Guardrails

Enterprise-grade AI Brand Manager using Model Context Protocol (MCP), NVIDIA NIM guardrails, and social media integration.

🚀 Features

• Real MCP Protocol: Actual Model Context Protocol client-server communication
• Multi-layered Guardrails: NVIDIA NeMo Guardrails with Content Safety and Topic Control
• Social Media Integration: Automated Mastodon posting with safety validation
• Enterprise UI: Professional Gradio interface for brand management workflows
• GPU-accelerated: NVIDIA NIM containers for high-performance inference

🏗️ Architecture

UI → AI Agent → MCP Server → NeMo Guardrails → NVIDIA NIMs → Mastodon API

• UI Layer: Gradio-based professional interface
• Agent Layer: Strategic content planning and execution
• MCP Layer: Standards-compliant tool calling protocol
• Guardrails Layer: 23-category safety validation
• Infrastructure: Docker containers with GPU allocation

📋 Prerequisites

• NVIDIA GPU with Docker support
• NGC API Key from NVIDIA
• Mastodon account with API access
• Python 3.8+
• Docker with NVIDIA Container Runtime

🛠️ Setup Instructions

1. Clone the Repository

git clone ssh://[email protected]:12051/skrithivasan/guardrailsmcp.git
cd guardrailsmcp

2. Environment Configuration

cp .env.example .env
# Edit .env with your actual API keys

3. Install Python Dependencies

pip install -r requirements.txt

4. Start NVIDIA NIM Services

# Set up NIM cache
export LOCAL_NIM_CACHE=~/.cache/nim
mkdir -p "${LOCAL_NIM_CACHE}"

# Create Docker network
docker network create nemoguard-local

# Start LLM service (GPU 0)
docker run --rm -d --network=nemoguard-local --name=llm \
  --gpus="device=0" --runtime=nvidia \
  -e NVIDIA_API_KEY \
  -u $(id -u) \
  -v "${LOCAL_NIM_CACHE}:/opt/nim/.cache" \
  -p 8000:8000 \
  nvcr.io/nim/meta/llama-3.1-8b-instruct:1.8.6

# Start Content Safety service (GPU 1)
docker run --rm -d --network=nemoguard-local --name=contentsafety \
  --gpus="device=1" --runtime=nvidia \
  -e NVIDIA_API_KEY \
  -u $(id -u) \
  -v "${LOCAL_NIM_CACHE}:/opt/nim/.cache" \
  -p 8001:8000 \
  nvcr.io/nim/meta/llama-3.1-nemoguard-8b-content-safety:1.0.0

# Start Topic Control service (GPU 2)
docker run --rm -d --network=nemoguard-local --name=topiccontrol \
  --gpus="device=2" --runtime=nvidia \
  -e NVIDIA_API_KEY \
  -u $(id -u) \
  -v "${LOCAL_NIM_CACHE}:/opt/nim/.cache" \
  -p 8002:8000 \
  nvcr.io/nim/meta/llama-3.1-nemoguard-8b-topic-control:1.0.0

5. Start NeMo Guardrails

docker run --rm -d --network=nemoguard-local --name=guardrails \
  --gpus=all --runtime=nvidia \
  -e NVIDIA_API_KEY \
  -v $(pwd)/config-store:/opt/app/config-store \
  -p 7331:8000 \
  nvcr.io/nim/nemo-guardrails:24.08

6. Launch the Application

# Start the AI Brand Manager UI
python3 real_mcp_ui.py

🔧 Usage

1. Access the UI: Open the Gradio interface (typically http://localhost:7862)
2. Enter Task: Describe your brand management task
3. Review Process: Watch real-time execution with safety checks
4. Approve Content: Review generated content and guardrails analysis
5. Social Publishing: Content is automatically posted to Mastodon if safe

Example Tasks

• "Announce our new AI product launch"
• "Create a customer appreciation post"
• "Share company quarterly results"
• "Promote our upcoming webinar"

🛡️ Safety Features

Content Safety Categories (S1-S23)

• Violence, Sexual content, Criminal planning
• Weapons, Controlled substances, Self-harm
• Minor safety, Hate speech, PII/Privacy
• Harassment, Threats, Profanity
• Copyright/Trademark violations
• Unauthorized advice, Illegal activities

Topic Control Guidelines

• Business announcements ✅
• Technology discussions ✅
• Customer appreciation ✅
• Offensive content ❌
• Spam/promotional abuse ❌

📊 Key Files

• real_mcp_ui.py: Main Gradio UI application
• mastodon_mcp_server.py: MCP server for social media tools
• config-store/nemoguard/: Guardrails configuration
• requirements.txt: Python dependencies
• technical_flow_explanation.md: Detailed architecture documentation

🔍 Troubleshooting

Common Issues

1. GPU Memory: Ensure sufficient GPU memory for 3 NIM containers
2. API Keys: Verify NVIDIA_API_KEY is valid and exported
3. Network: Check Docker network connectivity between services
4. Ports: Ensure ports 7331, 8000-8002 are available

Debugging Commands

# Check container status
docker ps

# View container logs
docker logs llm
docker logs guardrails

# Test API endpoints
curl http://localhost:7331/v1/health
curl http://localhost:8000/v1/health

📈 Performance

• End-to-end Workflow: ~4-6 seconds
• Safety Validation: ~2-3 seconds (parallel GPU processing)
• Social Media Posting: ~1-2 seconds
• MCP Protocol Overhead: ~0.1-0.3 seconds

🤝 Contributing

This project demonstrates enterprise AI workflows with:

• Standards-compliant MCP protocol implementation
• Production-ready safety validation
• Real-time social media integration
• Professional UI/UX design

📝 License

NVIDIA Internal Project