# computer MCP server

> Enables secure macOS computer use for AI agents with automatic password redaction, user consent for writes, and comprehensive audit logging.

- Repository: https://github.com/Agent360dk/computerMCP
- Page: https://hostedmcp.org/servers/agent360dk/computermcp
- Transport: stdio
- Stars: 1
- License: MIT
- Latest version: 0.1.0

## Install

### Claude Code

```sh
claude mcp add computermcp -- npx -y @agent360/computer-mcp
```

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

```json
{
  "mcpServers": {
    "computermcp": {
      "command": "npx",
      "args": [
        "-y",
        "@agent360/computer-mcp"
      ]
    }
  }
}
```

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

```json
{
  "servers": {
    "computermcp": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@agent360/computer-mcp"
      ]
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "computermcp": {
      "command": "npx",
      "args": [
        "-y",
        "@agent360/computer-mcp"
      ]
    }
  }
}
```

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

```toml
[mcp_servers.computermcp]
command = "npx"
args = ["-y", "@agent360/computer-mcp"]
```

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

```json
{
  "mcpServers": {
    "computermcp": {
      "command": "npx",
      "args": [
        "-y",
        "@agent360/computer-mcp"
      ]
    }
  }
}
```

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `CMCP_ASK_TIMEOUT` | no | no | Seconds a consent dialog waits before refusing. Default 60. |
| `CMCP_MODE` | no | no | readonly \| ask (default) \| allow |

## FAQ

### Does it protect my passwords?

Yes, secure text fields and password manager windows are painted opaque black in memory before any screenshots are written to disk.

### Can I run multiple agents at once?

Yes, the server supports multiple simultaneous sessions without a session lock.

### How does it handle Electron apps like VS Code or Slack?

It actively switches on the accessibility tree for Electron apps so they are not seen as empty windows by the agent.

### Is there an audit trail of what the agent does?

Every action is recorded in an append-only, fingerprinted log that includes the target app and the decision outcome.