memory-mcp-server

A FastMCP server that bridges Claude Code and Claude.ai by exposing Claude Code project memory files over MCP.

Problem

Claude Code and Claude.ai are separate products with separate contexts. Claude Code stores project memory as markdown files locally under ~/.claude/projects/*/memory/. Claude.ai has no access to these files, so you lose context every time you switch between the two.

Related MCP server: Cortex

Solution

This server exposes those memory files via MCP over HTTPS, so Claude.ai can read and write them through a connector — giving both interfaces shared memory across all projects.

Architecture

Claude.ai └─→ Cloudflare Tunnel (https://memory.yourdomain.com/mcp) └─→ FastMCP server (Ubuntu VM) └─→ ~/.claude/projects//memory/.md ↕ rsync every 5 min WSL ~/.claude/projects//memory/.md ↑ Claude Code

MCP Tools

Setup

1. Server (Ubuntu VM)

git clone https://github.com/Ahmedaltu/memory-mcp-server.git
cd memory-mcp-server
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python server.py

2. Systemd service

sudo cp memory-mcp.service /etc/systemd/system/
sudo systemctl enable memory-mcp
sudo systemctl start memory-mcp

3. Cloudflare Tunnel

cloudflared tunnel login
cloudflared tunnel create memory-mcp
cloudflared tunnel route dns memory-mcp memory.yourdomain.com
sudo cloudflared service install
sudo systemctl enable cloudflared
sudo systemctl start cloudflared

4. WSL auto-sync (add to crontab -e)

# Push WSL → VM (every 5 min)
*/5 * * * * rsync -az -e "ssh -i ~/.ssh/your-key.pem" --include="*/" --include="*/memory/" --include="*/memory/*.md" --exclude="*" ~/.claude/projects/ ubuntu@YOUR_VM_IP:~/.claude/projects/

# Pull VM → WSL (every 5 min)
*/5 * * * * rsync -az -e "ssh -i ~/.ssh/your-key.pem" --include="*/" --include="*/memory/" --include="*/memory/*.md" --exclude="*" ubuntu@YOUR_VM_IP:~/.claude/projects/ ~/.claude/projects/

5. Connect Claude.ai

Go to Settings → Connectors → Add connector and enter: https://memory.yourdomain.com/mcp

License

MIT