Skip to main content
Agent setup

Give your AI agents a safe, local memory.

Clypy serves MCP itself. Turn it on in Settings, connect Claude, Codex, Cursor and more with one click, and your agents can recall safe clips and save what you ask them to remember. Nothing else to install. Part of Clypy Pro on Mac, Windows and Linux.

Local data path
Your AI agent
Clypy, in MCP mode
The Clypy app on this computer
Encrypted local library

No public endpoint. The agent starts Clypy's own app in MCP mode, which talks only to Clypy running on the same computer, and starts it if it isn't.

Nothing else to install

Agents run the Clypy app you already have, so the connection updates with it, and no secrets go into agent config files.

Hard exclusions

Concealed, password and expiring clips are removed from every agent result, and an agent can't edit or delete them either.

Your switches

Off until you turn it on. Choose whether agents can read full text and save or change clips, and remove an agent any time.

Quick start

Connect once, use from every project

01

Turn it on

In Clypy, open Settings › MCP and switch on Let AI agents use Clypy. MCP is part of Clypy Pro; the same page unlocks it.

02

Connect your agents

Clypy lists the agents it finds on your computer. Click Connect next to each one, then restart it so it loads Clypy. Clypy adds one entry to the agent's own settings and leaves everything else there exactly as it was; Remove takes it out again.

Claude CodeClaude desktop appCodexCursorGemini CLIWindsurfVS CodeKimi Code

Codex runs searches and new saves without asking and still asks before it edits or deletes anything. Other agents ask the way they ask for any tool.

03

Any other agent: paste a prompt

On the same page, Copy a setup prompt and paste it into any agent that can edit its own settings. It carries the exact path to Clypy on your computer, and the agent adds Clypy to its own MCP configuration, then asks you to restart it.

To do it by hand instead, add a stdio MCP server named clypy. Set up by hand shows the entry with your path filled in:

Most MCP clients
{
  "mcpServers": {
    "clypy": {
      "command": "<path to Clypy, shown in Settings › MCP>",
      "args": ["mcp"]
    }
  }
}

Some clients call the containing field mcpServers, servers, or mcp_servers. The command and arguments stay the same.

04

Optional: add the Clypy Memory skill

The skill teaches agents when to save, to search before saving, and to update a memory instead of duplicating it. The Skills CLI puts it where your installed agents look:

Clypy Memory skill
npx skills add https://updates.clypy.app/skills/clypy-memory-skill.zip

Add --global to make it available across your projects. Review skills before installing them: a skill only teaches the workflow, and the MCP connection provides the tools.

Using memory

Collections are choices, not sandboxes

An agent can keep its own memory collection when that is useful, but it can also work in collections you already use with people and other agents. What it saves syncs to your devices like anything you copy, marked with the agent that saved it.

Own collection
“Save this in Codex Memory.”
Shared collection
“Save this in Project Phoenix.”
Several collections
“Save this in Research and Team Decisions.”
Unfiled
“Remember this, but leave it outside collections.”
Keep it current
“The deadline moved to Friday; update what you saved.”
Secret boundary

Some memory is deliberately unreadable.

  • Concealed, password and expiring clips never appear in agent results, and agents can't change them.
  • A clip an agent saves as sensitive is concealed before it is ever written or synced.
  • Collection membership never overrides the exclusion.
  • Save a reference to the password manager, not the credential itself.

Agent can't see Clypy?

Restart the agent after connecting it, and check Settings › MCP shows it as Connected with the switch on. An agent that says MCP is turned off or needs Pro is passing on Clypy's own answer.