AI Sandbox & MCP Integration Guide

Embedenv now provides a unified runtime infrastructure for AI agents and LLMs. Connecting Claude Desktop, Cursor, or your autonomous python backend to isolated sandbox containers takes just minutes using either standard Model Context Protocol (MCP) or our REST APIs.

1. Cloud MCP Server Setup (Secure Cloud Containers)

The **Cloud MCP Server** runs inside a secure, isolated cloud container. AI agents connect to this server over HTTP (via Server-Sent Events or direct POST requests). All files are managed and code is executed on the cloud server, meaning your local machine does not require any code runtimes installed.

Claude Desktop Configuration

Add the following block to your claude_desktop_config.json file:

{
  "mcpServers": {
    "embedenv-cloud-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/client-sse",
        "https://embedenv.com/api/mcp/sse?token=YOUR_PUBLIC_API_KEY"
      ]
    }
  }
}

Cursor IDE Configuration

Go to **Cursor Settings -> Features -> MCP** and click **Add New MCP Server**:

• Name: Embedenv-Cloud-MCP
• Type: SSE
• URL: https://embedenv.com/api/mcp/sse?token=YOUR_PUBLIC_API_KEY

2. Local MCP Server Setup (Local Machine Execution)

The **Local MCP Server** runs directly on your computer via stdio communication. It has direct access to your local files, local terminal shell, web browser, and screen. It is distributed as a ZIP bundle containing all scripts.

Claude Desktop Configuration

Add the following block to your claude_desktop_config.json file:

{
  "mcpServers": {
    "embedenv-local-mcp": {
      "command": "python",
      "args": [
        "C:/path/to/extracted/distributable_mcp/mcp_server_client.py",
        "--public-key",
        "YOUR_PUBLIC_API_KEY",
        "--skills",
        "files,shell,web,screen",
        "--backend",
        "https://embedenv.com"
      ]
    }
  }
}

Cursor IDE Configuration

Go to **Cursor Settings -> Features -> MCP** and click **Add New MCP Server**:

• Name: Embedenv-Local-MCP
• Type: command
• Command: python -u C:/path/to/extracted/distributable_mcp/mcp_server_client.py --public-key YOUR_PUBLIC_API_KEY --skills files,shell,web,screen --backend https://embedenv.com

3. Differences & Use Cases

Here is a detailed comparison of the two MCP server architectures offered by Embedenv:

Cloud MCP Use Cases

• Low-Power Local Systems: AI can run heavy compilers, dependencies, databases, and visual runtimes in the cloud workspace without stressing your laptop.
• Remote/Web IDEs: Connecting browser-based agents or cloud workspaces (like Gitpod, GitHub Codespaces, Replit) directly.
• Untrusted Source Code: Safely testing unknown packages, parsing risky files, or running code written by AI without risk to your local system's data and network.

Local MCP Use Cases

• Local Software Engineering: Directly modifying local git repos, compiling code on your system toolchain, or updating files in VS Code/Cursor.
• Interactive GUI Testing: Taking real screenshots of your actual primary display monitor to audit local web apps or desktop layouts.
• Offline Operations & Intranet: Executing commands that interact with local network databases, smart home nodes, or local docker clusters.

4. Available MCP Tools

The AI model gains access to the following workspace tools. The **Cloud MCP** server supports all 12 tools (both cloud-native and local-parity naming conventions), while the **Local MCP** server exposes the local-specific subset.

3. REST Sandbox API (Ephemeral)

For custom autonomous loops where you manage tasks via API keys, call our endpoints directly:

A. Create / Allocate Sandbox

curl -X POST https://embedenv.com/api/v1/sandbox/create/ \
  -H "Authorization: Bearer YOUR_PUBLIC_API_KEY"

{
  "ok": true,
  "sandbox_id": "sbx-a3df8b29c",
  "status": "building",
  "ws_url": "wss://embedenv.com/api/v1/sandbox/ws/sbx-a3df8b29c",
  "message": "Sandbox workspace duplication initiated."
}

B. Execute Command / Code

curl -X POST https://embedenv.com/api/v1/sandbox/execute/ \
  -H "Authorization: Bearer YOUR_PUBLIC_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "sandbox_id": "sbx-a3df8b29c",
    "language": "python",
    "code": "import sys; print(\"Python running: \", sys.version)"
  }'

{
  "ok": true,
  "stdout": "Python running: 3.10.12 (main, Jun 11 2023, 05:26:28)\n",
  "stderr": "",
  "credits_remaining": 994
}

Live Sandbox & MCP Connection Monitor

This panel displays your active real-time connections. Values update automatically every few seconds.

Active MCP Server Connections

Active Ephemeral AI Sandboxes