BRIEFMCP

AI-powered knowledge base that gives every AI agent on your team the same, up-to-date project context—without copy-pasting massive markdown files.

📋 Devpost – Full hackathon pitch, solution context & screenshots

🏗️ Project Ecosystem

BRIEFMCP consists of three integrated repositories:

1. Brief Main Frontend (This Repo)

Repository: martinbon39/parisinnovhack

The main knowledge base platform with:

• Document management dashboard
• PDF upload and text extraction
• Multi-tenant architecture
• Token savings analytics
• Beautiful, responsive UI

2. BriefKit - Context-First Development Toolkit

Repository: marlowetal653/BriefKit

Constitution-driven and spec-driven development framework:

• /brief.context - Check project knowledge before any action
• /brief.constitution - Create project principles and governance
• /brief.specify - Define what to build with context awareness
• /brief.plan - Create technical plans using existing patterns
• Enforces John Lam's constitution-driven development methodology

3. parisinnov-mcp - MCP Server Package

Repository: martinbon39/parisinnov-mcp

Published npm package providing MCP tools:

• search-context - AI-powered semantic search (GLM-4.7)
• add-knowledge - Document ingestion
• list-knowledge - Knowledge base management
• delete-knowledge - Document removal
• Zero-install via npx parisinnov-mcp

🎯 The Problem

Teams waste 2+ hours per week searching for information and 88% of AI tokens by copy-pasting entire documents into prompts. Different AI agents make inconsistent decisions because they don't share the same knowledge base.

💡 The Solution

BRIEFMCP indexes your project's constitution and specifications into an optimized knowledge base, then uses RAG (Retrieval Augmented Generation) to retrieve only the most relevant sections. AI agents get exact constraints, requirements, and decisions—with citations—under a token budget.

Key Benefits:

• 🚀 88% token reduction - Retrieve 200 tokens instead of 10,000
• ⚡ < 2 seconds search response time
• 🔒 Multi-tenant - Secure, isolated workspaces per organization
• 🔌 Zero-install - Works with Claude, Cursor, GitHub Copilot via MCP

✨ Features

• Semantic Search - Find answers by meaning, not keywords
• PDF Upload - Automatic text extraction and indexing
• MCP Integration - Works with any MCP-compatible AI assistant
• Token Savings Tracking - Analytics dashboard showing cost savings
• Real-time Ingestion - Documents searchable within seconds
• Context-First Development - BriefKit integration for AI agents

🛠️ Tech Stack

Frontend:

• Next.js 16 (App Router)
• React 19
• Tailwind CSS
• Framer Motion

Backend:

• Supabase (PostgreSQL + pgvector)
• Edge Functions (Deno)
• Row Level Security (RLS)

AI/ML:

• Featherless AI GLM-4.7 (query understanding)
• Featherless AI BAAI/bge-large-en-v1.5 (embeddings)
• OpenAI text-embedding-3-small (embeddings)

Infrastructure:

• Vercel (frontend)
• Supabase (backend)
• Railway (MCP backend)

🚀 Quick Start

Prerequisites

• Node.js 18+
• npm or yarn
• Supabase account
• OpenAI API key (for embeddings)
• Featherless AI API key (optional, for client-side embeddings)

Installation

# Clone the repository
git clone https://github.com/martinbon39/parisinnovhack.git
cd parisinnovhack

# Install dependencies
npm install

# Set up environment variables
cp .env.example .env.local
# Edit .env.local with your credentials

# Run development server
npm run dev

Open http://localhost:3000 to see the app.

Environment Variables

NEXT_PUBLIC_SUPABASE_URL=your_supabase_url
NEXT_PUBLIC_SUPABASE_ANON_KEY=your_supabase_anon_key
FEATHERLESS_API_KEY=your_featherless_key  # Optional

Using the MCP Server

The MCP server is published as an npm package:

npx parisinnov-mcp --access-token YOUR_TOKEN

Configure it in your AI assistant (Claude, Cursor, etc.) to enable automatic context retrieval.

📁 Project Structure

parisinnovhack/
├── src/
│   ├── app/              # Next.js pages and routes
│   │   ├── dashboard/    # Main dashboard
│   │   ├── savings/      # Token savings analytics
│   │   └── auth/         # Authentication
│   ├── services/         # Business logic
│   │   ├── ingestion/    # Document ingestion pipeline
│   │   └── savings/      # Token savings calculations
│   └── lib/              # Utilities and helpers
├── supabase/
│   └── functions/        # Edge Functions
│       └── ingest-context/ # Document ingestion endpoint
└── scripts/              # Test data and utilities

🏗️ Architecture

Document Ingestion Flow:

1. PDF/Markdown upload → PDF.js extraction
2. Markdown-aware chunking (~1000 tokens/chunk)
3. Dual-path embedding generation (Featherless/OpenAI)
4. Vector storage in Supabase pgvector

Query Processing:

1. User query → MCP backend
2. GLM-4.7 understands intent
3. Vector similarity search
4. Contextual answer with citations

📊 Current Status

✅ MVP Complete

• Core document management
• PDF upload and processing
• Vector embeddings and semantic search
• MCP server published (parisinnov-mcp v1.1.0)
• Multi-tenant architecture
• Token savings tracking

🚧 In Progress

• In-app search UI
• Project-level document scoping
• Source integrations (Notion, Confluence)

🤝 Contributing

This project was built for the Paris Innov Hackathon. Contributions welcome!

📄 License

MIT License - see LICENSE file for details

🔗 Links

Hackathon Submission

• BRIEF MCP on Devpost – Full context, problem/solution narrative, technical stack, limits & roadmap, live demo link

Repositories

• Brief Main Frontend: martinbon39/parisinnovhack (This repo)
• BriefKit: marlowetal653/BriefKit
• MCP Server: martinbon39/parisinnov-mcp

Documentation

• Local Setup & Project Analysis: LOCAL_SETUP.md – internal: auth troubleshooting, env vars, base schema, build mode, what's missing
• MCP Package: npm - parisinnov-mcp
• Pitch Document: PITCH.md
• MCP Usage Guide: MCP_USAGE_GUIDE.md
• Constitution & Spec-Driven Design: CONSTITUTION_SPEC_DRIVEN_DESIGN.md
• Core Value Proposition: CORE_VALUE_PROPOSITION.md

---

Built with ❤️ for the Paris Innov Hackathon