Connect your AI

With local MCP, AI agents on your computer, such as Claude, Codex or Cursor, can search the Uni3K cards you allow and save notes in a space of their own. Everything passes between the browser and the agent on your own computer.

Set it up in a minute

Install Node.js 18 or later if you don't have it yet (nodejs.org, LTS version).
In Uni3K, open Settings → AI connections and turn on Local MCP. Choose Allow when the browser asks.
Name your AI, for example Claude, and choose Connect AI.
Copy the message and send it to your AI. It downloads a small helper, registers it and adds Uni3K to itself.
Keep the browser open. The status in Settings shows when the helper is connected.

The message contains a token for that AI only and is shown once. The AI follows this guide, written for agents.

What your AI can do

Tool	What it does
`search_cards`	Search by keyword, with optional folder, site and date filters
`get_card`	Read one card in full
`list_folders`	List your folders and the agent's own space
`save_card`	Save a new Markdown card in the agent's own space
`update_card`, `delete_card`	Change or delete the agent's own cards; deleted cards go to Trash

You stay in control

An agent sees only its own space until you turn on Can read my cards for it.
Locked cards are never sent to an agent. It sees only that a locked card exists.
An agent can change or delete only the cards it wrote. You see them as Agent cards.
New token replaces an agent's token. Remove cuts it off and moves its cards to Trash.
Uni3K stores only a hash of each token. Nothing goes through our servers.

Set it up by hand

Prefer to do it yourself? Download uni3k-mcp.mjs (one file, no dependencies, you can read it first) and register it. On macOS or Linux:

mkdir -p "$HOME/.uni3k/mcp"
curl -fsSL https://uni3k.com/uni3k-mcp.mjs -o "$HOME/.uni3k/mcp/uni3k-mcp.mjs"
node "$HOME/.uni3k/mcp/uni3k-mcp.mjs" install

On Windows (PowerShell):

New-Item -ItemType Directory -Force "$env:USERPROFILE\.uni3k\mcp" | Out-Null
Invoke-WebRequest https://uni3k.com/uni3k-mcp.mjs -OutFile "$env:USERPROFILE\.uni3k\mcp\uni3k-mcp.mjs"
node "$env:USERPROFILE\.uni3k\mcp\uni3k-mcp.mjs" install

Then choose Connect AI in Settings, open Set it up by hand and paste the settings into your AI. They look like this:

{
  "mcpServers": {
    "uni3k": {
      "command": "node",
      "args": ["/Users/you/.uni3k/mcp/uni3k-mcp.mjs"],
      "env": { "UNI3K_TOKEN": "u3k_…" }
    }
  }
}

In Claude Code, one command does it:

claude mcp add uni3k --scope user --env UNI3K_TOKEN=u3k_… -- node "$HOME/.uni3k/mcp/uni3k-mcp.mjs"

Use the full path to your home folder in config files, because ~ and $HOME are not expanded there. To remove the helper, run it with uninstall.

If it doesn't connect

The agent says Uni3K is not reachable.

Keep Chrome or Edge open with Uni3K installed, check that Local MCP is on in Settings → AI connections, and that you chose Allow in the browser prompt. Then ask the agent to try again.

The agent says the token is unknown.

The token was replaced or the agent was removed. Choose New token for that agent and give your AI the new message.

node is not found.

Install Node.js 18 or later. If a desktop app still cannot find it, use the full path from which node (Windows: where node) as the command.