Sandbox

You run and connect to an MCP server that powers enhanced Python sandbox execution, artifact management, interactive features, and optional web app hosting. You can access it locally via HTTP or run it directly as a local stdio server, enabling integration with editors and AI tooling for rapid experimentation and automation.

How to use

Use the MCP server by connecting your client or editor to either the HTTP endpoint or the local stdio interface. The HTTP mode exposes a remote MCP URL you can reach from web or desktop apps, while the stdio mode lets you launch the server locally and communicate through the standard input/output streams. Typical workflows include executing Python code, capturing artifacts like plots and images, running Manim animations, and launching web apps from within your development environment.

How to install

{
  "mcpServers": {
    "sandbox_http": {
      "type": "http",
      "name": "sandbox_http",
      "url": "http://localhost:8765/mcp",
      "args": []
    }
  }
}

{
  "mcpServers": {
    "sandbox_stdio": {
      "type": "stdio",
      "name": "sandbox_stdio",
      "command": "sandbox-server-stdio",
      "args": [],
      "env": {}
    }
  }
}

Prerequisites you need before starting:
- Python 3.9+
- uvx or pip as your package manager
- Optional: a Python virtual environment tool available on your system
Then you can start either the HTTP MCP server or the stdio MCP server using the commands below.

Configuration and usage notes

HTTP mode: Run the HTTP server and connect your MCP-enabled client to the provided URL. The server discovers and exposes endpoints for tool execution and artifact retrieval, enabling automation and integration with editors or IDEs.

Stdio mode: Start the local server and interact via standard input/output. This mode is ideal for direct integration with development environments that can spawn and communicate with a background process.

Security and safety

The server implements command filtering to block dangerous operations, isolates execution in sandbox environments, and enforces configurable timeouts and resource monitoring. You can rely on these safeguards to run user-provided code with reduced risk to your system.

Examples of common tasks

Execute Python code and capture artifacts such as plots or images. Run web apps using Flask or Streamlit. Create Manim animations with built-in presets. Start an interactive REPL for exploration. All tasks return structured results and accessible artifact listings.

Notes on MCP tooling and integration

The server provides a set of MCP tools for code execution, artifact management, and environment inspection. You can list, run, and clean artifacts, start interactive sessions, and generate or manage animations programmatically through your MCP client.

Troubleshooting tips

If you encounter connection issues, verify that the HTTP URL is reachable or that the stdio process is running correctly. Check that the appropriate port (for HTTP) or the correct stdio command is used, and ensure your client is configured to communicate with the chosen transport.

Notes on examples and further usage

You can adapt the provided examples to your own code, artefact generation, and animation workflows. The server is designed to be extended with additional tools and workflows as your use cases grow.

Appendix: MCP JSON connection snippets

{
  "mcpServers": {
    "sandbox_http": {
      "type": "http",
      "name": "sandbox_http",
      "url": "http://localhost:8765/mcp",
      "args": []
    },
    "sandbox_stdio": {
      "type": "stdio",
      "name": "sandbox_stdio",
      "command": "sandbox-server-stdio",
      "args": [],
      "env": {}
    }
  }
}

Security & Safety details

The sandboxed execution environment ensures isolation, with configurable limits to prevent abuse. You can tune timeouts and resource usage to align with your needs and security requirements.

Available tools

execute

Execute Python code with artifact capture

shell_execute

Safely execute shell commands with security filtering

list_artifacts

List generated artifacts from executions

cleanup_artifacts

Remove temporary artifacts and clean up workspace

get_execution_info

Retrieve environment diagnostics and execution context

start_repl

Launch an interactive Python REPL session

start_web_app

Launch Flask or Streamlit web applications and expose endpoints

cleanup_temp_artifacts

Maintain artifact storage by removing temporary files

create_manim_animation

Create Manim animations programmatically and render outputs

list_manim_animations

List available Manim animation artifacts and templates

cleanup_manim_animation

Clean up specific Manim animation artifacts

get_manim_examples

Provide example code snippets for Manim demonstrations