HOPX

The HOPX MCP Server enables your AI assistants to execute code in isolated cloud containers through the Model Context Protocol.

It acts as a safe “playground” for your AI that spins up in milliseconds, runs the requested task (data analysis, file manipulation, or bash commands), and destroys itself immediately after.

Features

• 🐍 Multi-language support (Python, JavaScript, Bash, Go)
• 🔒 Isolated cloud containers with auto-cleanup
• 📊 Data science libraries pre-installed (pandas, numpy, matplotlib)
• ⚡ Fast container startup (~200ms)
• 🔄 Persistent sandbox environments
• 🌐 Internet access enabled
• 📁 Full file system operations
• ⏱️ Background and async task support

Use Cases

• Data analysis workflows where you need to process CSV files or API data without local Python setup.
• Rapid prototyping of scripts and algorithms through AI assistants.
• Educational scenarios where students can run code safely without local installations.
• Automated testing of code snippets across different languages.
• Processing sensitive data in isolated environments rather than locally.

How to Use It

1. Go to hopx.ai and generate your HOPX_API_KEY from the dashboard.

2. Install the MCP server via uvx.

uvx hopx-mcp

3. Add the server definition to your specific tool’s configuration file. Replace your-api-key-here with the actual key you generated.

For Claude Desktop

{
  "mcpServers": {
    "hopx-sandbox": {
      "command": "uvx",
      "args": ["hopx-mcp"],
      "env": {
        "HOPX_API_KEY": "your-api-key-here"
      }
    }
  }
}

For Cursor

{
  "mcpServers": {
    "hopx-sandbox": {
      "command": "uvx",
      "args": ["hopx-mcp"],
      "env": {
        "HOPX_API_KEY": "your-api-key-here"
      }
    }
  }
}

4. Once connected, the LLM will have access to a suite of tools. You can prompt the model naturally (e.g., “Run this python script to calculate Fibonacci”), but understanding the underlying tools helps you debug.

One-Shot Execution (Best for simple scripts)

The execute_code_isolated tool spins up a container, runs the code, and kills the container immediately. It’s the most cost-effective method for quick questions.

• Parameters: code, language (Python, JavaScript, Bash, Go), timeout.
• Example Output:

{
    "stdout": "Hello, World!\n",
    "exit_code": 0,
    "execution_time": 0.123
}

Persistent Sandboxes (Best for complex tasks)

If your agent needs to write a file in step 1 and read it in step 2, it needs a persistent session. The agent will use these tools in sequence:

1. create_sandbox(template_id='code-interpreter'): Establishes the environment.
2. execute_code(sandbox_id, code='...'): Runs code in that specific environment.
3. file_write(sandbox_id, path, content): Saves data to the container.
4. delete_sandbox(sandbox_id): Cleans up resources.

File System Operations

The agent can manipulate the container’s file system using:

• file_read / file_write
• file_list / file_exists
• file_mkdir / file_remove

FAQs

Q: Is the code running on my local machine?
A: No. All code executes in HOPX’s cloud infrastructure. Your local machine only acts as the client sending the request.

Q: Can the sandboxes access the internet?
A: Yes, internet access is enabled by default. This allows the code to pip install libraries, curl websites, or interact with public APIs.

Q: What happens if the AI forgets to delete a sandbox?
A: There is an auto-cleanup mechanism. By default, sandboxes self-destruct after 600 seconds (10 minutes) of inactivity or after the configured timeout is reached, preventing resource leaks.

Q: Can I use custom Docker images?
A: Currently, you select from available templates (like code-interpreter). You can use list_templates to see what environment configurations are available.

Q: Is my data safe inside the container?
A: The containers are isolated from each other and protected by JWT authentication. However, since the code runs in a cloud environment, you should adhere to your organization’s policy regarding uploading highly sensitive PII or credentials to third-party execution environments.