https://memex.mcp-hub.stream/mcpnoreply@notify.cloudflare.com — check spam too.1. Open the instructions field of your AI:
| Client | Location |
|---|---|
| claude.ai | Settings → Profile → "Personal preferences" (global) or Project → "Project instructions". Required — claude.ai currently does not inject MCP server instructions into the system prompt. |
| Claude Code | ~/.claude/CLAUDE.md — optional, server instructions already apply here automatically. |
| Cursor | .cursorrules in the project or Settings → Rules for AI. |
| ChatGPT | Settings → Personalization → Custom Instructions. |
2. Paste this snippet there — your AI will then recall before answering and store new facts immediately:
IMPORTANT: Memex (MCP) is my ONE persistent memory across ALL AI assistants and sessions. MEMORY PROTOCOL: 1. RECALL FIRST — before answering anything that may build on prior knowledge (my projects, people, preferences, decisions, past conversations), call recall_memory. 2. STORE PROACTIVELY — the moment I state durable information or ask you to remember something, call store_memory IMMEDIATELY. Never wait for the end of the conversation. 3. UPDATE, DON'T DUPLICATE — when a stored fact changes, use supersede_memory. ASSUME INTERRUPTION: unstored knowledge is permanently lost. Never store secrets or API keys.
1. "Remember: <new fact>" → does the AI store it immediately?
2. New session, ask about that fact → does it recall first?
3. "That fact changed: …" → does it update instead of duplicating?
Connection: https://memex.mcp-hub.stream/mcp · The memex_help tool lists everything else.