# Liquid Compute GPU test

This MCP server runs acceptance tests on NVIDIA GPU nodes and returns a report for the selected profile. Use it to check hardware before putting it to work.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI
- Author: Liquid-Compute
- License: MIT · policy=link-only · description only; the file stays with the author
- Source: https://github.com/Liquid-Compute/gcat
- Updated: 2026-10-15
- HTML: https://affpapa.org/en/nevibecode/mcp/liquid-compute-gpu-test

## What it does

The agent selects a profile for your nodes, helps arrange SSH access, and starts the test after a connection check. The report includes NVIDIA diagnostics and measurements of compute performance, memory bandwidth, and networking; results are compared with the profile thresholds. Profiles cover A10, H100, H200, B200, and B300 nodes, plus two- or four-node clusters using RDMA.

You need a Liquid Compute account and nodes running Ubuntu 22.04 or 24.04, NVIDIA driver 580 or later, and CUDA 13.0 or later. The nodes must accept SSH from the service's addresses, and the user needs root or passwordless sudo. Keep every GPU idle during the test. It puts the GPUs under full load and may run for 90 minutes or longer, so occupied GPUs rule out a safe run.

The agent must get your separate approval before adding nodes and starting a test; turn off automatic approval for these actions. Temporary test files are removed afterwards, but SSH access remains until you delete the line from `authorized_keys`. The report describes measurements from the test window; it is neither certification nor a warranty.

## Requirements
- Remote server: `https://api.liquidcompute.co/mcp`

## Install

### Claude Code (native platform)

**Platform CLI: remote server**
```bash
claude mcp add --transport http gcat https://api.liquidcompute.co/mcp
```

### Codex

**Platform CLI: remote server**
```bash
codex mcp add liquid-compute-gpu-test --url https://api.liquidcompute.co/mcp
```

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "liquid-compute-gpu-test": {
            "url": "https://api.liquidcompute.co/mcp"
        }
    }
}
```

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "liquid-compute-gpu-test": {
            "type": "http",
            "url": "https://api.liquidcompute.co/mcp"
        }
    }
}
```

### Gemini CLI

**Platform CLI: remote server**
```bash
gemini mcp add -t http liquid-compute-gpu-test https://api.liquidcompute.co/mcp
```

## When to use it
- check an NVIDIA GPU node against a hardware profile
- compare node measurements with profile thresholds
- verify SSH connectivity before loading the GPUs
- review hardware measurements from a test run

## Example prompts
- Run an acceptance test on my NVIDIA GPU nodes and select a suitable profile.
- Check SSH access to the nodes, then wait for my approval before starting the test.
- Show me the test results and prepare a report link.

## Similar
- [prisma-compute](https://affpapa.org/en/nevibecode/skills/prisma-compute.md) — skill · 371800 installs
- [gke-compute-classes](https://affpapa.org/en/nevibecode/skills/gke-compute-classes.md) — skill · 6508 installs
- [aws-compute](https://affpapa.org/en/nevibecode/skills/aws-compute.md) — skill · 5500 installs
- [gke-ai-troubleshooting-handle-disruption-gpu-tpu](https://affpapa.org/en/nevibecode/skills/gke-ai-troubleshooting-handle-disruption-gpu-tpu.md) — skill · 4700 installs
- [shapely-compute](https://affpapa.org/en/nevibecode/skills/shapely-compute.md) — skill · 495 installs
- [pint-compute](https://affpapa.org/en/nevibecode/skills/pint-compute.md) — skill · 482 installs

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md