Cross-model memory
for people and teams.

1. Open Settings → Connectors.
2. Click Add custom connector.
3. Paste the URL https://mcp.nuramem.ai/mcp and click Add.
4. Click Connect and sign in. Nuramem's tools now appear in any chat.

1. Open Settings → Connectors.
2. Click Add custom connector.
3. Paste the URL https://mcp.nuramem.ai/mcp and click Add.
4. Click Connect and sign in. The connector works on claude.ai in the browser and in the Claude mobile apps.

Add the server from your terminal, then authenticate inside Claude Code:

claude mcp add --transport http --scope user nuramem https://mcp.nuramem.ai/mcp

The --scope user flag (a command-line flag — it goes before the server name) makes Nuramem available across all your projects; omit it to add it to the current project only. Then open Claude Code, run /mcp, select nuramem, and complete the browser sign-in.

Requires ChatGPT Plus (or Business/Enterprise).

1. Open Settings → Apps & Connectors.
2. Under Advanced, enable Developer mode.
3. Choose Create / Add custom connector, set the URL to https://mcp.nuramem.ai/mcp and Authentication to OAuth.
4. Create, then sign in. Nuramem is now available in chats.

Add Nuramem to ~/.gemini/settings.json (or a project .gemini/settings.json):

{
  "mcpServers": {
    "nuramem": {
      "httpUrl": "https://mcp.nuramem.ai/mcp",
      "oauth": { "enabled": true }
    }
  }
}

Restart Gemini CLI; it will prompt you to authenticate on first use. (Google's consumer Gemini app doesn't support custom MCP servers yet — use the Gemini CLI or the REST API for now.)

All connect to the same server. The fastest way is the CLI — it writes the right config for each, safely:

curl -fsSL https://get.nuramem.ai | sh   # installs `nura`
nura connect                              # detects + configures your clients

Or add it by hand: in Cursor and VS Code, add an MCP server pointing at https://mcp.nuramem.ai/mcp; in Windsurf, set serverUrl to the same. Restart the client and complete the browser sign-in on first use.

Save and recall from the terminal, CI, or any script — the same memory your assistants read.

curl -fsSL https://get.nuramem.ai | sh        # or: uv tool install nuramem · pipx install nuramem
nura login                                    # sign in (opens your browser)
nura this "we picked Postgres for the data plane"
nura search "data plane"                       # also: nura briefing · nura --json …

On macOS/Linux brew install nuramem/tap/nura and on Windows scoop install nura / winget install Nuramem.Nura also work.

Teach Claude Code, Cursor or Codex to reach for your memory reliably (the gesture, proactive capture, session-start briefing). The skill ships inside the CLI:

nura skill install                 # → ~/.claude/skills/nuramem/ (Claude Code)
nura skill install --client cursor # or cursor · codex · windsurf

Then restart the assistant. (To install by hand instead, copy the nuramem/ skill folder into the assistant's skills directory.)

MCP clients load Nuramem's tool list once, when the connection starts, and cache it for that session — so a newly added tool won't appear in a client you already had open. You don't need to remove and re-add the connector: just start a new chat or restart the client and it re-fetches the current tools. (Refreshing one app doesn't update another — each client keeps its own session.)

For MCP clients, no — you only ever paste the URL, and sign-in is OAuth (Nuramem never sees your password). For calling the REST API directly, you authenticate with a bearer token: sign in to the management app and copy it from Developers → API access.

In Nuramem's managed databases on Google Cloud — encrypted in transit and at rest, isolated per user, and built so saves and recalls are instant. Synthesizing your records into the compact picture your assistant loads is the product, and it requires our systems to read your memory content — so we hold it, and we protect it accordingly. We never sell it and never use it to train AI models.

You stay in control in two concrete ways: export everything anytime from the management app, and delete means deleted — deleting your account physically erases your records, typically within a minute, with residual copies aging out of encrypted backups within seven days. Details in the Privacy Policy.

Any MCP-compatible client. We've verified Claude Desktop, Claude Code, ChatGPT (Developer mode) and the Gemini CLI; other MCP clients — Cursor, Windsurf, Zed and more — connect the same way, to the same server and the same memory. Programmatic clients and agents can use the REST API at https://nuramem.ai/openapi.json.

The habit is naming Nura (or Nuramem) plus what you want: "nura, save this", "nura, remember the deadline moved to Friday", "nuramem, what's my latest decision on billing?", "nura, move these to the Acme project." Naming Nura is the part that matters — it ensures the memory lands in your cross-model layer instead of the assistant's own store. Nuramem also captures the durable things you mention on its own and leaves the chit-chat alone.

It saves typed records — not chat logs — and at the start of a session hands the model a compact, current picture of who you are and what you're working on, on whichever model you're using.

No — and that's new. Nuramem now captures proactively: when you mention something durable — a decision, a deadline, a person and their role, a reference you're leaning on, a lesson learned, a change in your status — it saves it to the right place and tells you where, without being asked. When something's tentative or ambiguous it checks first, and it ignores passing chit-chat.

You're always in control: you can say "don't save that," browse and delete anything from the management app, and export or erase everything whenever you choose.

Yes — that's shared projects, now available. Create a project in the management app (or just ask your assistant to "create a shared project"), then invite teammates by email and give each a role. They accept the invitation from the app the next time they sign in.

Once they're in, the decisions, plans and context recorded in that project are recalled by every member's assistant — on whichever model each person uses. Everyone keeps their own private personal memory; only what you put in the project is shared. And if two people record conflicting decisions, Nuramem surfaces both — with who decided what — rather than silently picking a winner.

Nuramem is currently free — every feature, including shared projects. We expect to introduce paid plans later; if we do, we'll give notice before anything is charged. And you're never locked in: export your entire memory anytime.