Persistent memory for Claude Desktop with Slowave

Give Claude Desktop a local memory that carries across conversations. Slowave connects over MCP; you paste one block into Custom Instructions.

Slowave gives Claude Desktop access to the same local memory as your other agents. Setup needs one manual paste.

What you get

• Setup adds the Slowave server to Claude Desktop’s config file.
• Claude Desktop starts Slowave over stdio. This differs from the other agents.
• You paste Slowave’s instructions into Custom Instructions yourself.
• Memory stays on your machine.

Set up

1. Install Slowave

   pipx install slowave
   
2. Preview and run setup

   slowave setup --client claude-desktop --dry-run
   slowave setup --client claude-desktop
   

   Setup prints Slowave’s instruction block in your terminal.

3. Paste the instructions

   In Claude Desktop, open Settings, then General, then Instructions for Claude. Paste the block that setup printed.

4. Restart Claude Desktop

   Quit and reopen the app.

Custom Instructions apply before the first turn. Skills may load later, so they can’t replace this step. Claude Desktop stores the field on Anthropic’s side, so setup can’t fill it in.

Check that it works

Start a new conversation. The slowave_* tools should appear. Then run:

slowave doctor

Good to know

• Setup detects Claude Desktop through its application support folder. On macOS that is ~/Library/Application Support/Claude/.
• The config file is claude_desktop_config.json in that folder. Windows uses %APPDATA%\Claude\. Linux uses ~/.config/Claude/.
• Slowave stores memory locally. When recalled, that context goes to Claude under Anthropic’s data handling.

Troubleshooting

The tools don't appear

Check the config path shown by slowave setup --dry-run. Check that the JSON is valid, then restart Claude Desktop.

Memory only starts on turn two

The instructions are probably missing from Custom Instructions. Paste the block there instead of using a Skill.

Setup didn't connect the server

Add this to the config file, then restart. Find the path with which slowave-mcp on macOS or Linux:

{
  "mcpServers": {
    "slowave": {
      "command": "/path/to/slowave-mcp"
    }
  }
}

Quickstart

Install Slowave and see your first memory.

All integrations

Compare every supported agent.