arrakis

Arrakis MCP Server exposes the Arrakis VM sandbox functionality to Large Language Models via a set of MCP resources and tools. It provides a namespace arrakis:// for interacting with the available VMs and performing common VM lifecycle operations. Use the resources to discover and inspect VMs, and leverage the tools to start, snapshot, restore, run commands inside, upload/download files, and manage VM lifecycle states. This enables safe VM sandbox execution and snapshot-based workflows within your LLM-powered automation pipelines. The tools are designed to be invoked through MCP calls that specify the corresponding tool name and required parameters, returning structured results about VM state and I/O. If you need to manage multiple VMs or reproduce exact environments, the snapshot and restore features are particularly useful for creating and reusing VM images.

Prerequisites:

• Python 3.8+ installed on the host
• A working installation of uv (the UV runtime) for Python-based MCP servers
• Access to a repository containing the Arrakis MCP server script (arrakis_mcp_server.py) and its dependencies

Installation steps:

1. Install uv if you don’t already have it

   # Example (install via pip).
   python -m pip install uv
   

2. Clone the Arrakis MCP server repository

   git clone https://github.com/abshkbh/arrakis-mcp-server.git
   cd arrakis-mcp-server
   

3. Ensure you have Arrakis-related dependencies available in your environment (the server will import arrakis_mcp_server.py from the repository path and use it to manage VMs).

4. Run the MCP server via UV with the provided configuration

   # Adjust <path-to-uv> and <path-to-repo> to your environment
   uv --directory <path-to-repo> run arrakis_mcp_server.py
   

5. Verify the server is responding (optional): you can probe the MCP endpoint using your usual MCP client tooling to list resources or invoke a tool like start_sandbox.

Note: If you plan to run multiple MCP servers or need environment customization, consider exporting environment variables or updating the arrakis_mcp_server.py as needed to point to the correct Arrakis VM sandbox backend.

Tips and common issues:

• Ensure <path-to-repo> is the directory containing arrakis_mcp_server.py and its dependencies; the --directory flag tells UV where to locate the script.
• If you encounter module import errors, verify that Python can access the repository path and that all Arrakis-specific dependencies are installed.
• The MCP namespace for this server is arrakis:// and supports resources like arrakis://vms and arrakis://vm/{vm_name}, as well as tools such as start_sandbox, restore_snapshot, snapshot, run_command, upload_file, download_file, destroy_vm, destroy_all_vms, and update_vm_state.
• When deploying in production, consider pinning a specific Python/UV version and isolating the environment to avoid compatibility issues with the Arrakis VM sandbox.
• When using start_sandbox or run_command, provide sufficient VM configuration (VM name, sandbox settings) as required by arrakis_mcp_server.py.