# CozoDB Memory MCP server

> Provides AI agents with a local-first, persistent memory system using a unified engine for relational, graph, vector, and full-text search.

- Repository: https://github.com/tobs-code/cozo-memory
- Page: https://hostedmcp.org/servers/tobs-code/cozo-memory
- Transport: stdio
- Stars: 1
- License: Apache-2.0
- Latest version: 1.0.3

## Install

### Claude Code

```sh
claude mcp add cozo-memory -- npx -y cozo-memory
```

### Cursor (`~/.cursor/mcp.json`)

```json
{
  "mcpServers": {
    "cozo-memory": {
      "command": "npx",
      "args": [
        "-y",
        "cozo-memory"
      ]
    }
  }
}
```

### VS Code (`.vscode/mcp.json`)

```json
{
  "servers": {
    "cozo-memory": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "cozo-memory"
      ]
    }
  }
}
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "cozo-memory": {
      "command": "npx",
      "args": [
        "-y",
        "cozo-memory"
      ]
    }
  }
}
```

### Codex (`~/.codex/config.toml`)

```toml
[mcp_servers.cozo-memory]
command = "npx"
args = ["-y", "cozo-memory"]
```

### Gemini CLI (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "cozo-memory": {
      "command": "npx",
      "args": [
        "-y",
        "cozo-memory"
      ]
    }
  }
}
```

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `EMBEDDING_MODEL` | no | no |  |

## Tools (5)

- `mutate_memory`
- `query_memory`
- `analyze_graph`
- `manage_system`
- `edit_user_profile`

## FAQ

### What are the hardware requirements for running the default model?

You need a minimum of 1.7 GB of RAM to run the default bge-m3 model.

### Can I use a more lightweight model if I have low RAM?

Yes, you can use the Xenova/all-MiniLM-L6-v2 model, which only requires approximately 400 MB of RAM.