How to Add Persistent Memory to Claude.ai (5-Minute Setup)

Give Claude a brain that persists across sessions. Self-hosted, open source, works directly in your browser — no desktop app required.

The Problem

Claude is powerful, but it doesn't remember context between sessions. Every conversation starts from scratch.

Sure, ChatGPT has memory. Gemini has memory. Claude Desktop has extensions.

But what if you:

• Use claude.ai in the browser (no desktop app)
• Want to own your data (no third-party cloud)
• Need enterprise security (OAuth 2.0, HTTPS, CORS)
• Require self-hosted infrastructure (compliance, privacy)

That's where mcp-memory-service comes in.

What is mcp-memory-service?

An open-source persistent memory layer for Claude that implements the Model Context Protocol (MCP). Unlike other solutions, it supports Remote MCP — meaning it works directly in claude.ai browser without requiring Claude Desktop.

What Makes It Different?

5-Minute Setup Tutorial

Let's get mcp-memory-service running with Remote MCP for claude.ai.

Prerequisites

• Python 3.10+
• A machine that can run 24/7 (VPS, home server, or your laptop for testing)
• 5 minutes of your time

No domain name required — we'll use Cloudflare Tunnel for free HTTPS.

# Clone the repository
git clone https://github.com/doobidoo/mcp-memory-service.git
cd mcp-memory-service

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Linux/Mac
# or: venv\Scripts\activate  # Windows

# Install dependencies
pip install -e .

# Remote MCP Configuration
MCP_STREAMABLE_HTTP_MODE=1
MCP_SSE_HOST=0.0.0.0
MCP_SSE_PORT=8765

# OAuth Configuration
MCP_OAUTH_ENABLED=true
MCP_OAUTH_STORAGE_BACKEND=sqlite
MCP_OAUTH_SQLITE_PATH=./data/oauth.db

# Memory Configuration (optional)
MCP_MEMORY_CONSOLIDATION_ENABLED=true
MCP_MEMORY_DECAY_ENABLED=true

python -m mcp_memory_service.server

[INFO] MCP Memory Service starting...
[INFO] Streamable HTTP transport enabled
[INFO] OAuth 2.0 server initialized
[INFO] Server running on http://0.0.0.0:8765
[INFO] MCP endpoint: http://0.0.0.0:8765/mcp

# macOS
brew install cloudflared

# Linux
wget https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64.deb
sudo dpkg -i cloudflared-linux-amd64.deb

# Start tunnel
cloudflared tunnel --url http://localhost:8765

Your quick Tunnel has been created! Visit it at:
https://random-name-1234.trycloudflare.com

You: "Store a memory: My favorite programming language is Rust"
Claude: "I've stored that memory."

--- close conversation, open new one ---

You: "What's my favorite programming language?"
Claude: "Your favorite programming language is Rust!"

How It Works

1. You interact with Claude in the browser
2. Claude detects memory tools from your connector
3. When relevant, Claude calls memory_store() or memory_search()
4. Request goes to your server via HTTPS (OAuth token authenticates it)
5. Memory is stored/retrieved from your local database
6. Claude uses the context in its response

All communication is encrypted (HTTPS), authenticated (OAuth 2.0), and private (data never leaves your infrastructure).

The Knowledge Graph

Unlike simple key-value storage, mcp-memory-service builds a knowledge graph with typed edges:

Memory: "Favorite language is Rust"
  ↓ relates_to: programming
  ↓ preference_for: language
Memory: "Working on a web server"
  ↓ uses_technology: backend
  ↓ relates_to: architecture

This allows Claude to find related memories contextually, understand relationships between facts, and provide better answers by connecting the dots.

Auto-Consolidation

Over time, memories accumulate. mcp-memory-service automatically:

• Decays — Reduces importance of old, unused memories
• Compresses — Merges similar memories (e.g., 5 mentions of "likes Rust" → 1 consolidated memory)
• Prunes — Archives memories with zero relevance score

Production Deployment

Option 1: Permanent Cloudflare Tunnel

# Authenticate
cloudflared tunnel login

# Create tunnel
cloudflared tunnel create memory-service

# Route traffic
cloudflared tunnel route dns memory-service memory.yourdomain.com

# Run (use systemd/supervisor in production)
cloudflared tunnel run memory-service

Option 2: Your Own Domain + Let's Encrypt

# Get certificate
certbot certonly --standalone -d memory.yourdomain.com

# Update .env
MCP_HTTPS_ENABLED=true
MCP_SSL_CERT_FILE=/etc/letsencrypt/live/memory.yourdomain.com/fullchain.pem
MCP_SSL_KEY_FILE=/etc/letsencrypt/live/memory.yourdomain.com/privkey.pem

# Restart server
python -m mcp_memory_service.server

Option 3: Reverse Proxy (Caddy)

The simplest production setup — Caddy handles HTTPS automatically:

memory.yourdomain.com {
    reverse_proxy localhost:8765
}

That's it. Caddy auto-provisions Let's Encrypt certificates.

Cost Analysis

Self-hosted on a Hetzner CX21 (2 vCPU, 4GB RAM) at ~$6/month. DigitalOcean ~$12, AWS t3.small ~$15.

Real-World Use Cases

Monitoring & Maintenance

# Health check
curl https://your-domain.com/api/health

# Database backup
cp ./data/memories.db ./backups/memories-$(date +%Y%m%d).db

# Server logs (systemd)
journalctl -u mcp-memory-service -f

Troubleshooting

Conclusion

Adding persistent memory to Claude doesn't have to be complicated or expensive.

With mcp-memory-service and Remote MCP, you get:

• Browser-based workflow (no desktop app)
• Enterprise security (OAuth 2.0, HTTPS)
• Full data ownership (self-hosted)
• Advanced features (knowledge graph, auto-consolidation, hybrid search)
• Cost-effective ($6/month vs $50+/month SaaS)

Setup takes 5 minutes. The value lasts forever.

Resources

• mcp-memory-service on GitHub
• Official MCP Documentation
• Anthropic: Building Custom Connectors
• Cloudflare Tunnel Documentation
• MCP Remote Server Submission Guide

About the Author — Henry (doobidoo) is a Senior Technical Consultant and DevOps Engineer specializing in cloud infrastructure and AI tooling. He built mcp-memory-service out of frustration with vendor lock-in and lack of browser-based memory solutions for Claude.

Connect on GitHub.