srunx MCP server

srunx is a helper tool for people who run jobs on a SLURM cluster, which is the shared computer setup many universities and labs use for heavy computing work. It lets you submit jobs, check the queue, and run multi-step pipelines from one place instead of juggling several commands. This page is about the MCP server part, which lets your AI assistant do those things for you.

What this MCP server does

You ask your AI something like submit this job or run my workflow. The AI passes your request to the srunx MCP server. The server then talks to your SLURM cluster, either on your own machine or over SSH to a remote one. It submits the job, checks its status, or starts the workflow, and sends the result back. You see the answer right in the chat, like a job id or a status update.

Click to zoom

What you can do with it

• Submit a SLURM job to your cluster
• Check the queue and see which jobs are running
• Run a YAML workflow file with its steps and dependencies
• Run a parameter sweep across several values at once
• Target a remote cluster by naming a saved SSH profile
• Look at job logs and status changes

Try asking your AI

• “Submit my train.sh script to the cluster with 2 GPUs per node”
• “What jobs are running in the queue right now?”
• “Run my workflow.yaml file and tell me when it finishes”
• “Run train.yaml as a sweep over lr values 0.001 and 0.01, two at a time”

What it gives back to you

You get answers in plain text in the chat. That could be a job id after submitting, a list of jobs and their states, or a note that a workflow started or finished. If something fails, the server tells you what went wrong. You do not get files or charts, just the information and confirmation of what it did.

Before you start

Who it's for

Researchers, data scientists, and lab or university staff who run jobs on a SLURM cluster and want their AI assistant to handle the day-to-day submitting and checking.