Skip to content
MCP server

NVSentinel Lab

MCP server for testing NVIDIA NVSentinel GPU fault handling on a kind lab cluster

Runs locally
Install
Pricing
Open source
Platforms
CLI, Linux, macOS
Runs
Locally
Version
0.5.1
Listing updated
today

Similar MCP servers

About NVSentinel Lab

NVSentinel Lab is an open-source MCP server for cloud and infrastructure. It runs locally on your machine and is started by your AI client.

The README describes this as a Go-based MCP server that lets an MCP client read what NVSentinel reports on a Kubernetes lab, inject an XID 95 fault into a simulated GPU, and check that NVSentinel cordons and drains the node. It does not install NVSentinel itself; that setup is handled by separate repositories the author links to.

The server exposes read tools for node state, scenarios, component logs and events, plus change tools (fault injection, clear, reset and end-to-end scenario runs). Changing tools are disabled by default and only work when explicitly enabled and limited to a kind cluster context whose name starts with kind-.

The README says it is aimed at people testing NVSentinel's detection and remediation pipeline in a lab. It states the results prove the pipeline against simulated DCGM, not real GPU health.

Install

Run in a terminal

claude mcp add nvidia-nvsentinel-mcp -- docker run -i --rm docker.io/omerfsen/nvidia-nvsentinel-mcp:0.5.1

Add to claude_desktop_config.json

{
  "mcpServers": {
    "nvidia-nvsentinel-mcp": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "docker.io/omerfsen/nvidia-nvsentinel-mcp:0.5.1"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "nvidia-nvsentinel-mcp": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "docker.io/omerfsen/nvidia-nvsentinel-mcp:0.5.1"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "nvidia-nvsentinel-mcp": {
      "type": "stdio",
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "docker.io/omerfsen/nvidia-nvsentinel-mcp:0.5.1"
      ]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "nvidia-nvsentinel-mcp": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "docker.io/omerfsen/nvidia-nvsentinel-mcp:0.5.1"
      ]
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "nvidia-nvsentinel-mcp": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "docker.io/omerfsen/nvidia-nvsentinel-mcp:0.5.1"
      ]
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

Configuration

  • NVSENTINEL_MCP_KUBECONFIGThe one kubeconfig it uses, private to you. Default in the image: /kube/config.
  • NVSENTINEL_MCP_ALLOW_CHANGESExactly true allows the change tools, on a kind cluster only. Off when unset.
  • NVSENTINEL_MCP_LAB_CONTEXTThe kubeconfig context changes need. Must start with kind-. Default kind-nvsentinel-lab.

Tools this server provides (9)

Taken from the project documentation; the running server may differ.

lab_status
Nodes (cordon, NVSentinel state, failing GPU checks), pods, the workload
list_scenarios
Scenarios, stages, allowed XIDs, log sources
wait_for_stage
Polls until a stage holds, such as cordoned
component_logs
The last lines of a NVSentinel component's logs
node_events
Recent events on the GPU node
inject_fault
Injects XID 95 into the simulated GPU
clear_fault
Restarts the simulated hostengine, which drops the fault
reset_lab
Undoes a DCGM outage and an injected fault
run_test_scenario
Runs a scenario end to end and reports each step

What it does

Read cluster state
Reports node cordon status, NVSentinel state, failing GPU checks, pods and the workload.
Inject simulated GPU fault
Injects XID 95 into the simulated GPU when change tools are enabled.
Wait for pipeline stage
Polls until NVSentinel reaches a stage such as cordoned.
Predefined test scenarios
Runs the xid95 and dcgm_outage scenarios end to end and reports each step.
Component logs and events
Returns recent NVSentinel component log lines and events on the GPU node.
Lab-only safety guards
Change tools stay off unless explicitly allowed and only work on kind contexts.

Good for

  • Testing NVSentinel fault detection and cordon behaviour on a kind cluster
  • Checking that a workload is evicted and drained after a simulated GPU fault
  • Letting an AI coding client inspect NVSentinel state during lab debugging
  • Reproducing a DCGM outage and confirming the monitor reports it

Pricing

Open source

Open-source server; no paid plan mentioned.

NVSentinel Lab FAQ

Does this server install NVSentinel?

No. The README says it does not install NVSentinel; that is handled by separate linked repositories.

Can it change a real cluster?

Change tools are off unless NVSENTINEL_MCP_ALLOW_CHANGES is set to exactly true, and they are meant to work only on a kind cluster.

Does it test real GPU health?

The README says its test results prove NVSentinel's pipeline against simulated DCGM, not real GPU health, and a real GPU has not been checked.

How is it installed?

The README lists release binaries, a go install command and a Docker image, and shows how to add it to Claude Code with claude mcp add.

Is NVSentinel Lab free?

Yes. NVSentinel Lab is open source, so you can use and self-host it for free.

How do I add NVSentinel Lab to Claude Code?

Run this command in your terminal: claude mcp add nvidia-nvsentinel-mcp -- docker run -i --rm docker.io/omerfsen/nvidia-nvsentinel-mcp:0.5.1

What are the best alternatives to NVSentinel Lab?

The closest alternatives to NVSentinel Lab are Liquid Compute GPU test, XportL, Proxmox VE MCP Server, EMT Compute and MCP Canary.

Where this comes from

  • Description: written from the project's README and its MCP Registry entry.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers