Memxus: A secure memory layer for Claude, ChatGPT, Cursor and Telegram

Every day I switch between Claude, ChatGPT and Cursor. And every day I re-explain the same things: my stack, my preferences, my project context.

After calculating I was losing 30–60 minutes daily doing this — 150+ hours per year — I built Memxus.

The problem

AI tools don't talk to each other. Every new session starts from zero. You end up being the human copy-paste layer between your own tools.

Existing solutions either work for one platform only, require browser extensions with questionable permissions, or are built for agents — not for developers switching between tools all day.

What Memxus does

Save a memory once in any AI tool. Recall it instantly in all the others.

In Claude: remember("My stack is Next.js + Supabase + Railway. Always use TypeScript.")

Later, in Cursor, ChatGPT, VSCode, Telegram: recall("my stack") → "Next.js + Supabase + Railway. Always use TypeScript."

No copy-paste. No re-explaining. Your context travels with you.

Where it works today

• Claude (web, mobile, desktop) — MCP OAuth
• ChatGPT — OAuth Actions (GPT Store)
• Cursor / VS Code — MCP Bearer
• Gemini web — Chrome extension
• Gemini CLI — MCP OAuth
• Telegram — @memxus_bot
• Discord — connected

Security first

This handles your private context, so security was non-negotiable:

• OAuth 2.1 PKCE — industry standard auth, no passwords stored
• Encrypted at rest — all memories use AES encryption (mxe1: format)
• Row-level security — complete isolation between users and teams via Supabase RLS
• RFC 9728 compliant — follows the official MCP standard
• Portable data — export or delete everything, anytime

Teams and groups

Memxus also supports shared memory for teams. A group shares context across members — each member's private memories stay private, shared memories are accessible to the group only.

Pricing: Free / Pro $12/mo / Team $49/mo

The MCP angle

MCP (Model Context Protocol) is becoming the universal standard for AI tool integrations. Memxus is built on top of it as infrastructure — not a feature of one tool, but a layer that any MCP-compatible client can connect to.

This means every new AI tool that adopts MCP becomes a potential integration with zero additional work.

Try it

→ memxus.com — free to start, no credit card required

Would love feedback from the dev community — especially on the security model, what integrations you'd want next, and how you're managing context across AI tools today.