Give Claude & Cursor persistent memory in 5 minutes

By the end of this tutorial, Claude and Cursor will remember what you tell them — across new chats, across projects, and even across the two apps. No SDK, no glue code, about five minutes.

Large language models forget everything the moment a conversation ends. The context window is a whiteboard that gets wiped between sessions. This tutorial gives your assistant a memory that persists: you connect it to Aether once over the Model Context Protocol (MCP), and from then on it can save facts to — and recall them from — a durable knowledge base that outlives any single chat.

The proof is the last step: you'll tell your assistant something, close the chat and open a brand-new one, and watch it recall what you said.

The 5-minute path

1. Create a free Aether account — 30 seconds, no credit card.
2. Connect your assistant — Claude (one click) or Cursor (paste one config block).
3. Prove it remembers — teach it a fact, start a fresh session, ask again.

Let's go.

Step 1 — Create your free Aether account

You need an Aether account so your assistant has somewhere to store its memory. It's free and takes under a minute.

• Connecting Cursor or Claude Code? You'll also need an API key. Mint one on the API Keys page. The dashboard's Install MCP server card can generate a key scoped for MCP and drop it straight into the config snippet for you — grab your tenant id from the same screen while you're there.
• Connecting a Claude app (claude.ai / desktop)? You don't need a key at all — you'll sign in through your browser in the next step.

Step 2 — Connect your assistant

Pick the one you use. You only need one to see the memory work; wiring up both is what lets memory travel between them.

Option A — Claude (claude.ai, desktop, or Cowork)

Claude connects to Aether's hosted connector. There's no key to paste — you sign in and approve access once.

1. In Claude, open Settings → Connectors.

2. Click Add custom connector.

3. Paste the Aether connector URL:

   text

   CopyCopied!

   https://mcp.aetherdb.ai/mcp
   

4. Confirm. Claude hands you to Aether's sign-in page — sign in (or create your account from Step 1), pick the workspace, and approve read + write access.

Claude can now search and update your Aether knowledge base from any conversation. Full walkthrough: Connect Aether to Claude.

Option B — Cursor

Cursor launches the local Aether MCP server with npx. Add this to ~/.cursor/mcp.json (global) or a project .cursor/mcp.json, using the API key and tenant id from Step 1:

{
  "mcpServers": {
    "aether": {
      "command": "npx",
      "args": ["-y", "@aether-ai/mcp-server"],
      "env": {
        "AETHER_API_KEY": "<your-aether-api-key>",
        "AETHER_TENANT_ID": "<your-tenant-id>"
      }
    }
  }
}

Restart Cursor. Under Settings → MCP, aether turns green once it connects. You'll need Node.js 18+ so npx can launch the server — nothing to install ahead of time. Full walkthrough: Model Context Protocol (MCP).

Also using Claude Code?

One line in your terminal wires it up the same way:

claude mcp add aether \
  --env AETHER_API_KEY=<your-aether-api-key> \
  --env AETHER_TENANT_ID=<your-tenant-id> \
  -- npx -y @aether-ai/mcp-server

Step 3 — Prove it remembers across sessions

This is the whole point. You'll save a fact, throw away the conversation, and recall it from a clean one.

1. Teach it something

In your assistant, say:

"Remember this in my Aether knowledge base: our production database is Postgres 16, and the read replica lives in us-east-1."

Behind the scenes the assistant calls the aether_insert tool (Cursor / Claude Code) or the connector's save tool (Claude app) and stores that sentence in your knowledge base. You'll usually see it confirm the save.

2. Throw the conversation away

Close the chat and start a brand-new one. In Cursor, open a fresh chat; in Claude, start a new conversation. This clears the context window — a normal assistant now knows nothing about your database.

3. Ask again — from the clean session

In the new conversation, ask:

"What do you know about my production database?"

The assistant calls aether_search (or aether_retrieve), finds the fact you saved earlier, and answers with your Postgres 16 / us-east-1 details — even though this conversation never heard them.

What just happened

The memory didn't live in the chat — it lived in Aether. That's the difference between conversation memory (a transcript that dies with the tab) and a persistent knowledge base (a corpus you build up once and reuse everywhere):

Because it's a real corpus, the same knowledge is available to any app you build on the Aether SDKs (Python, TypeScript, Go, .NET) — the assistant and your code read and write the exact same memory.

Troubleshooting

The assistant doesn't see the Aether tools. Restart the client fully after connecting — MCP servers and connectors are wired up at startup. For Cursor, check .cursor/mcp.json is valid JSON (a trailing comma is the usual culprit) and that aether shows green under Settings → MCP.

Cursor: npx: command not found or the server won't launch. Install Node.js 18+ and make sure node / npx are on your PATH. The first launch downloads @aether-ai/mcp-server, so give it a few seconds and stay online.

Authentication failed / 401 (Cursor / Claude Code). The AETHER_API_KEY is missing, mistyped, or revoked — mint a fresh one on the API Keys page and update the config.

More fixes on the MCP integration page.

Next steps

Your assistant remembers now. Build on it: