README.md

🚀 IntegraChat

Multi-Tenant Autonomous MCP Platform

Enterprise-grade AI with autonomous agents, secure multi-tenant RAG, real-time web search, red-flag governance, and analytics.

---

📑 Table of Contents

• Overview
• Purpose
• Key Features
• Technology Stack
• System Architecture
• Project Structure
• Getting Started
• Why IntegraChat Stands Out
• Submission Metadata
• License

---

📌 Overview

IntegraChat is an enterprise-ready, multi-tenant AI platform built to demonstrate the full capabilities of the Model Context Protocol (MCP) in a real production-style environment.

It combines autonomous tool-using agents, RAG retrieval, live web search, and admin governance under strict tenant isolation, powered by Ollama (local) or Groq (cloud) LLM inference.

IntegraChat is a complete "MCP in Action" ecosystem — ideal for enterprise demos, research, production scaffolds, and governance-focused AI deployments.

---

🎯 Purpose

IntegraChat showcases how MCP can power intelligent, governed, multi-tenant AI systems with real-world requirements:

• 🔒 Isolation & Access Control - Strict tenant separation
• 🛡️ Compliance & Red-Flag Detection - Automated safety monitoring
• 🤖 Tool-Aware Autonomous Reasoning - Dynamic tool selection
• 🔍 RAG + Web Search Hybrid AI - Best of both worlds
• 📊 Analytics & Observability - Full system insights
• ⚙️ Admin Governance Workflows - Enterprise-ready controls

---

🧩 Key Features

1. 🤖 Autonomous MCP Agents

Agents can intelligently:

• ✅ Analyze user intent and context
• ✅ Detect sensitive or unsafe content
• ✅ Dynamically select multiple tools in sequence (RAG + Web + LLM, Web + LLM, RAG + LLM, etc.)
• ✅ Multi-step tool execution - Execute tools sequentially and synthesize results
• ✅ Retrieve tenant-private knowledge
• ✅ Pull real-time data from the internet
• ✅ Trigger admin alerts when needed
• ✅ Respond using Ollama (local) or Groq (cloud) LLM

2. 📚 Enterprise RAG System

• MiniLM embeddings (384-dim) for semantic search
• Supabase + pgvector for vector storage
• Automatic database schema initialization on server startup
• Strict multi-tenant isolation with tenant_id filtering
• Intelligent text chunking (~300 words per chunk)
• Vector similarity search using cosine distance

3. 🌐 Live Web Search Tool

• Real-time news and information
• General web search capabilities
• English language results - Forces English content via region parameters
• Fact-checking & fresh data retrieval
• Multiple provider support (DuckDuckGo, SerpAPI, Bing)

4. 🔄 Multi-Tool Selection & Execution

The intelligent tool selector can:

• Pattern-based detection - Recognizes fact queries, freshness keywords, internal docs
• LLM-enhanced planning - Uses LLM to determine optimal tool combinations
• Sequential execution - Executes tools in order (RAG → Web → LLM)
• Result synthesis - Combines all tool outputs into comprehensive responses

Supported combinations:

• RAG + LLM - Internal knowledge questions
• Web + LLM - Public fact questions
• RAG + Web + LLM - Comprehensive queries needing both sources
• LLM only - Simple conversational queries

5. 🚨 Red-Flag Governance Engine

Admins configure rules to:

• ⛔ Block unsafe queries automatically
• 📝 Log violations for audit trails
• 🔔 Trigger admin alerts in real-time

Example rules: salary, delete all data, confidential client info

6. 📊 Analytics Dashboard

Comprehensive insights for:

• 📈 Query volume and trends
• 🔧 Tool usage statistics
• 🎯 RAG performance metrics
• 🚨 Red-flag violation tracking
• 🧠 Agent reasoning traces
• 👥 Tenant activity monitoring

7. 🏢 Multi-Tenant Isolation

Each tenant gets:

• 🔐 Private agents
• 📦 Private knowledge base
• ⚙️ Private admin rules
• 📊 Private analytics

Isolation is guaranteed via Supabase Row-Level Security (RLS).

---

🛠 Technology Stack

Backend

Technology	Purpose
FastAPI	High-performance API framework
MCP Client + Servers	Model Context Protocol implementation
Supabase	Auth + Storage + pgvector database
PostgreSQL + pgvector	Vector database for embeddings
MiniLM Embeddings	Semantic search embeddings (384-dim)
Ollama / Groq	LLM inference engine (configurable)
DuckDuckGo Search	Web search provider
Slack / Email	Alerting system

Frontend

Technology	Purpose
React 18	Modern UI framework
TailwindCSS	Utility-first styling
Zustand	State management
WebSocket	Real-time streaming
Admin Dashboard	Analytics & governance UI

---

🧱 System Architecture

┌─────────────────────────────────────────────────────────────┐
│                    Frontend (React)                         │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────┐  │
│  │ Chat UI  │  │  Admin   │  │Analytics │  │WebSocket │  │
│  │          │  │  Panel   │  │ Dashboard│  │Streaming │  │
│  └──────────┘  └──────────┘  └──────────┘  └──────────┘  │
└─────────────────────────────────────────────────────────────┘
                            ↕ HTTP/WebSocket
┌─────────────────────────────────────────────────────────────┐
│              MCP Client (FastAPI)                           │
│  • Intent handling                                          │
│  • Red-flag scanning                                        │
│  • Multi-tool selection logic                               │
│  • Sequential tool execution                                │
│  • Groq Llama-3.1 integration                              │
│  • Event logging                                            │
└─────────────────────────────────────────────────────────────┘
                            ↕ MCP Protocol
┌─────────────────────────────────────────────────────────────┐
│                    MCP Servers                              │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐     │
│  │ RAG Server   │  │ Web Search   │  │ Admin Server │     │
│  │ (Knowledge)  │  │ (Live/EN)    │  │ (Governance) │     │
│  └──────────────┘  └──────────────┘  └──────────────┘     │
└─────────────────────────────────────────────────────────────┘
                            ↕
┌─────────────────────────────────────────────────────────────┐
│              Supabase Database                              │
│  • Authentication                                           │
│  • pgvector (embeddings)                                    │
│  • Row-Level Security (RLS)                                 │
│  • Storage                                                  │
└─────────────────────────────────────────────────────────────┘

---

📁 Project Structure

IntegraChat/
├── backend/
│   ├── api/
│   │   ├── main.py
│   │   ├── routes/
│   │   │   ├── agent.py
│   │   │   ├── rag.py
│   │   │   ├── web.py
│   │   │   ├── admin.py
│   │   │   └── analytics.py
│   │   ├── services/
│   │   │   ├── agent_orchestrator.py
│   │   │   ├── intent_classifier.py
│   │   │   ├── redflag_detector.py
│   │   │   ├── tool_selector.py
│   │   │   ├── prompt_builder.py
│   │   │   └── llm_client.py
│   │   ├── mcp_clients/
│   │   │   ├── rag_client.py
│   │   │   ├── web_client.py
│   │   │   └── admin_client.py
│   │   ├── models/
│   │   │   ├── requests.py
│   │   │   ├── responses.py
│   │   │   ├── agent_decision.py
│   │   │   └── embeddings.py
│   │   ├── utils/
│   │   │   ├── logging.py
│   │   │   ├── supabase_client.py
│   │   │   ├── tenant_context.py
│   │   │   ├── security.py
│   │   │   └── text_extractor.py   
│   │   └── config.py
│   │
│   ├── mcp_servers/
│   │   ├── main.py              # RAG MCP Server (FastAPI)
│   │   ├── database.py          # Supabase/PostgreSQL connection
│   │   └── embeddings.py        # Sentence transformers embeddings
│   │
│   ├── workers/
│   │   ├── ingestion_worker.py
│   │   ├── analytics_worker.py
│   │   ├── scheduler.py
│   │   └── celeryconfig.py
│   │
│   ├── tests/
│   │   ├── test_agent.py
│   │   ├── test_rag.py
│   │   ├── test_admin.py
│   │   ├── test_web.py
│   │   └── test_end_to_end.py
│   │
│   └── Dockerfile
│
├── frontend/
│   ├── src/
│   │   ├── components/
│   │   │   ├── chat/
│   │   │   │   ├── ChatWindow.jsx
│   │   │   │   ├── MessageBubble.jsx
│   │   │   │   └── ToolTraces.jsx
│   │   │   ├── admin/
│   │   │   │   ├── RedFlagEditor.jsx
│   │   │   │   └── TenantManager.jsx
│   │   │   └── analytics/
│   │   │       ├── QueryStats.jsx
│   │   │       ├── ToolUsageChart.jsx
│   │   │       └── RagEffectiveness.jsx
│   │   ├── pages/
│   │   │   ├── ChatPage.jsx
│   │   │   ├── AdminPage.jsx
│   │   │   └── AnalyticsPage.jsx
│   │   ├── store/
│   │   │   ├── chatStore.js
│   │   │   ├── adminStore.js
│   │   │   └── analyticsStore.js
│   │   ├── utils/
│   │   │   ├── websocket.js
│   │   │   ├── api.js
│   │   │   └── formatters.js
│   │   ├── App.jsx
│   │   └── main.jsx
│   ├── public/
│   └── package.json
│
├── supabase/
│   ├── migrations/
│   │   ├── 001_init.sql
│   │   ├── 002_embeddings.sql
│   │   ├── 003_redflag_rules.sql
│   │   └── 004_analytics_tables.sql
│   ├── seed/
│   │   ├── default_redflags.sql
│   │   └── demo_tenants.sql
│   ├── policies/
│   │   ├── rls_tenant_documents.sql
│   │   ├── rls_tenant_embeddings.sql
│   │   └── rls_admin_rules.sql
│   ├── functions/
│   │   └── vector_search.sql
│   └── README.md
│
├── docs/
│   ├── architecture-diagram.png
│   ├── sequence-mcp-agent.png
│   ├── api-spec.md
│   ├── mcp-server-protocol.md
│   └── rag-pipeline.md
│
├── scripts/
│   ├── run_dev.sh
│   ├── build_all.sh
│   ├── reset_db.sh
│   └── run_workers.sh
│
├── docker-compose.yml
├── README.md
└── LICENSE

---

🚀 Getting Started

Prerequisites

Before you begin, ensure you have the following installed:

• ✅ Python 3.10+
• ✅ Node.js 18+ (for frontend)
• ✅ Supabase project (with pgvector extension enabled)
• ✅ PostgreSQL connection string (from Supabase)
• ✅ Ollama (for local LLM) or Groq API key (for cloud LLM)
• ✅ DuckDuckGo Search key (optional, if configured)
• ✅ Slack/Email webhook for alerts (optional)

Backend Setup

1. Navigate to backend directory

   cd backend
   

2. Install dependencies

   pip install -r requirements.txt
   

3. Configure environment variables

   Create a .env file in the project root with the following:

   # Database Configurationa
   POSTGRESQL_URL=postgresql://user:password@host:port/database
   SUPABASE_URL=https://your-project.supabase.co
   SUPABASE_SERVICE_KEY=your_service_role_key
   
   # MCP Server URLs
   RAG_MCP_URL=http://localhost:8001
   WEB_MCP_URL=http://localhost:8002
   ADMIN_MCP_URL=http://localhost:8003
   
   # LLM Configuration
   OLLAMA_URL=http://localhost:11434
   OLLAMA_MODEL=llama3
   # Or use Groq instead:
   # GROQ_API_KEY=your_groq_api_key
   

4. Start the RAG MCP Server

   cd backend/mcp_servers
   python main.py
   

   The server will automatically initialize the database schema on startup.

   • Server runs on http://localhost:8001
   • API docs available at http://localhost:8001/docs

5. Run the main API server

   cd backend
   uvicorn api.main:app --reload
   

RAG MCP Server API

The RAG MCP Server provides two main endpoints:

Ingest Documents:

curl -X POST http://localhost:8001/ingest \
  -H "Content-Type: application/json" \
  -d '{
    "tenant_id": "tenant123",
    "content": "Your document text here..."
  }'

Semantic Search:

curl -X POST http://localhost:8001/search \
  -H "Content-Type: application/json" \
  -d '{
    "tenant_id": "tenant123",
    "query": "What are the HR policies?"
  }'

Frontend Setup

1. Navigate to frontend directory

   cd frontend
   

2. Install dependencies

   npm install
   

3. Start the development server

   npm run dev
   

Quick Start with Docker

docker-compose up -d

---

✨ Why IntegraChat Stands Out

Feature	Description
🤖 True MCP-Native	Autonomous agents (not static prompts)
🛡️ Enterprise Governance	Regex-based red-flag rules system
🔍 Hybrid Intelligence	Multi-tool selection (RAG + Web + LLM combinations)
🔄 Sequential Execution	Execute multiple tools in sequence and synthesize results
🌐 English Web Search	Forces English language results for better accuracy
🏢 Production-Grade	Multi-tenant design with strict Supabase RLS
📊 Full Observability	Logs, analytics, tool events, violations
🎯 Demo-Ready	Perfect for enterprise presentations

---

🏷 Submission Metadata

Field	Value
Track	MCP in Action
Category	Enterprise
Tag	mcp-in-action-track-enterprise
Project Name	IntegraChat

Short Summary

IntegraChat is a multi-tenant AI platform where autonomous MCP-powered agents intelligently select and execute multiple tools in sequence (RAG + Web + LLM combinations), retrieve private knowledge using RAG, access live web information in English, and enforce admin-defined safety rules via a red-flag compliance system. It includes an analytics dashboard, advanced multi-tool selection engine, and strict tenant isolation.

---

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

📬 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

---

🙏 Acknowledgments

• Built with Model Context Protocol (MCP)
• Powered by Ollama or Groq for LLM inference
• Database powered by Supabase

---

Made with ❤️ for the MCP community

⬆ Back to Top