🖥️ MTCode GPU Server

Turn your gaming PC or workstation into a private GPU execution server. Remote users can run Python workloads on your computer in a restricted sandbox — without SSH access or public exposure, and with resource controls designed to reduce risk.

🛡 Privacy — End-to-end encrypted. No MTCode infrastructure in your data path.

⚡ Simplicity — No public IP, domain, VPN, or port forwarding.

🚀 Performance — Direct peer-to-peer data path. No relay hop.

Why It Exists

Access to GPU compute is often limited. Shared clusters in universities or research groups are frequently oversubscribed, with long queues and restricted availability — while many powerful GPUs in personal workstations, lab machines, and small team servers sit idle much of the time.

These GPUs are highly capable, but difficult to share safely. Traditional approaches like SSH expose the entire computer, making them risky for personal machines and awkward for temporary sharing between labs, collaborators, or research groups. Cloud services add cost and reduce control.

MTCode GPU Server provides a third option: controlled, sandboxed execution on hardware you own or administer. It can support personal GPU sharing, lab-to-lab collaboration, short-term borrowing, or limited-time rental-style access — without giving users full system access.

How It Works

MTCode GPU Server is built on the same core library used by the MTCode DirectLink platform, but operates as a fully standalone application — no separate installation of MTCode Server is required. It delivers the same core benefits: high performance, low latency, and strong data privacy, while keeping your GPU computer private and inaccessible from the public internet.

When a user connects through the MTCode RemoteGPU extension, the system establishes a direct peer-to-peer connection between the client and the GPU server. Data flows directly between the two endpoints and is end-to-end encrypted, without passing through an intermediate relay or platform server.

Once connected, remote users can upload Python projects, execute them on your GPU, and stream results back in real time — or run jobs offline and download outputs later.

All execution runs inside a restricted sandbox environment. Users can run workloads, but never gain direct access to your system.

Architecture showing the GPU server, RemoteGPU extension, and Platform Server interaction

How MTCode GPU Server, MTCode RemoteGPU extension, and Platform Server interact — direct data flows with end-to-end encryption

Key Features

Sandboxed Execution

All code runs in isolation. File access, system calls, and resource usage are tightly controlled.

Full GPU Utilization

Run PyTorch, TensorFlow, or custom CUDA workloads using your local GPU.

Real-Time Interaction

Stream logs, outputs, and results directly back to VS Code's Output terminal.

No SSH Required

Users run code—without gaining shell access or visibility into your computer.

No Public Exposure

No open ports, no public endpoints. Your computer stays private.

Resource Control

Limit disk usage, user file retention periods, and maximum execution time.

Why Not SSH?

SSH Access

Full system access. High risk. Requires trust and manual management.

MTCode GPU Server

Sandboxed execution only. Controlled, limited, and secure by design.

Security by Design

User code runs inside a restricted sandbox that blocks system commands, limits filesystem access, and prevents unauthorized operations.

Each user’s workspace and saved files are isolated from other users. The host computer’s files, processes, and system environment are protected by default.

Best Practice — Run GPU Server Under a Separate User Account

No sandbox can guarantee protection against every possible exploit. For personal use or trusted teams, the built-in sandbox provides strong protection for normal workloads. For untrusted or external users, we recommend running MTCode GPU Server under a separate, unprivileged operating-system account with access only to the directories and resources required for GPU jobs.

This adds an additional layer of defense: even if malicious code were to escape the application sandbox, it would still be limited by the operating system permissions of that account.

Getting Started

Download and install MTCode GPU Server on your GPU computer, then sign in with your administrator account.

MTCode GPU Server provides an easy-to-use graphical interface for managing remote GPU access. It automatically detects available Python interpreters and makes them selectable for remote users. You can add additional interpreters if needed, and configure shared datasets that all authorized users can access to avoid redundant dataset uploads.

Configure user access, disk quotas, user file retention periods, and maximum execution time. Each user can run one job at a time, whether online or offline.

Invited users create their own accounts through the link in the invitation email, then connect through the MTCode RemoteGPU extension in VS Code or through MTCode Studio. They can upload Python projects, run them on your GPU, stream results in real time, or submit offline jobs and retrieve outputs later.

MTCode GPU Server uses the same administrator and user account system as MTCode Server. If you already have an MTCode account, you can use the same credentials.

Want to see it in action? Watch the demo video to follow a remote user as they discover your server, connect, upload a Python project, and execute it — all from within VS Code or MTCode Studio.

System Requirements

Hardware

One or more powerful GPUs worthy as GPU server
Sufficient disk space

Software

Windows 10/11, or Linux
GPU toolkit installed (CUDA, ROCm, oneAPI, MUSA, or CANN)
Python interpreter(s) installed (Conda, venv, or system Python)
Machine learning frameworks such as PyTorch, TensorFlow, or Keras with GPU support are installed