Universal Memory MCP: Claude Code Hackathon 2nd Place

# Universal Memory MCP: Claude Code Hackathon 2nd Place

How a 3-hour Universal Memory MCP demo won 2nd place at the Claude Code Hackathon and became the foundation for Tvix agent memory.

JJoshua ParkNovember 3, 2025Aristo won 2nd place 🥈 at the Claude Code Builder Hackathon held in Seoul. We're thrilled to receive $6,000 in Claude API credits.

Click to enlargeCelebrating our 2nd place win at the Claude Code MCP Hackathon in Seoul ## The Hackathon: Next-Generation MCP Innovation

The finals challenged participants to build "Next-Generation MCP Innovation Products with Claude Code", leveraging Claude Code and MCP (Model Context Protocol) technology to develop creative and viable productivity solutions.

The event was judged by Benjamin Mann, co-founder of Anthropic, alongside other industry leaders, making it a unique opportunity to validate cutting-edge AI infrastructure ideas.

Why We Participated

We've been thinking about a fundamental problem in AI: how do you coordinate memory across multiple agents?

This hackathon was the perfect opportunity to prototype our vision using MCP.

What We Built: Universal Memory MCP

In just 3 hours, we built Universal Memory MCP, a server that creates a shared memory layer accessible to any MCP-compatible AI agent.

Watch our hackathon demo: Universal Memory MCP in action

What It Does

Memory Operations:

AI agents can add, search, and retrieve memories through simple MCP tool calls. All memories are stored in a centralized system with vector embeddings for semantic search.

Cross-Agent Sync:

When Claude stores a memory, Cursor can instantly retrieve it. When ChatGPT learns something about your project, all your other agents know it too.

Smart Retrieval:

Instead of dumping entire conversation histories, agents query only the relevant memories they need using semantic similarity search.

Our demo showed multiple AI agents collaborating on a task, sharing context in real-time without manual copy-pasting or repeated explanations.

Why This Matters

Today's reality: you use ChatGPT for research, Claude for writing, Cursor for coding. Each time you switch tools, you lose context and repeat yourself.

The problem is simple: Each agent operates in isolation. What one learns doesn't transfer to others.

Universal Memory MCP solves this by creating a unified memory layer at the protocol level. Now all your agents share the same context, automatically staying in sync without manual intervention.

What's Next: Building Tvix

We're turning Universal Memory MCP into Tvix, a production-ready unified memory layer for all your AI agents.

What We're Building

1. Universal Integration

Connect via MCP or Chrome Extension. Works with ChatGPT, Claude, Cursor, and all your existing AI tools.

2. One-Click Context Sync

Automatically pull in context from Notion, Slack, Google Drive, Gmail, and more. All your scattered information in one searchable memory base.

3. Multi-Agent Infrastructure

The memory coordination layer that makes real multi-agent services possible.

We're using the $6,000 in Claude API credits to push toward our public beta launch.

The prototype proved the same idea that now drives Tvix: agent memory should not be trapped inside one chat product or one local workspace. For the broader architecture, read Agent Memory Beyond RAG. For the product workflow this became, read how to build your second brain with Tvix.

What's Next

Tired of repeating yourself across AI tools? We're building Tvix to solve this.

The era of Prompt Engineering is ending. The era of Context Engineering is here.

Looking forward to sharing more updates as we continue building Tvix.