Add Memory to your AI agent over MCP

The @ragionex/memory-mcp server gives your AI agent a persistent memory it manages for you - it saves the facts worth keeping and recalls them when they are relevant, across sessions, projects, and tools. One key, one config block, the same memory everywhere.

Get a free key

Add the server (two ways)

A) Let your agent set it up. Paste this to your AI agent:

Add the @ragionex/memory-mcp MCP server to my config using my API key rgx_memory_...

B) Set it up yourself. Add this to your MCP client’s user-global config:

{
  "mcpServers": {
    "ragionex-memory-mcp": {
      "command": "npx",
      "args": ["-y", "@ragionex/memory-mcp"],
      "env": { "RAGIONEX_MEMORY_API_KEY": "rgx_memory_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx" }
    }
  }
}

The server runs locally via npx - nothing is installed globally, and -y always pulls the latest published version.

Start your agent

Start a fresh agent session. Tell it something worth keeping - a preference, a decision, a convention. It is there next session, and in every other tool you connect with the same key.

Supported clients

Because it speaks the Model Context Protocol, one server works across every MCP client. It is verified in:

Claude Code
Claude Desktop
Codex
ChatGPT (Apps SDK)
Cursor
Cline

…and 500+ MCP clients. Each client reads the same mcpServers object shape; they differ only in which file they store it in.

Use one key everywhere

Configure the same rgx_memory_ key in two tools and they share one memory store. Save in Cursor, recall in Claude Code - same key, same user, same memory.

Good to know

On startup the server also writes one short priority rule into your agent’s always-loaded rules file, so the guidance to reach for Memory first is never hidden or truncated. It is wrapped in markers, backs up your file before the first write, never touches your own content, and is fully reversible. See How the config rule works (and how to control it). A few clients (Claude Code, Codex, Windsurf) have quirks worth knowing - see MCP client notes and known limitations.

Next steps

Prefer calling the API directly? See Quickstart: Memory and the full Memory API reference.