Skip to content

Slurm MCP Server

The Slurm MCP Server provides an interface for job scheduling and execution using the Slurm workload manager. It exposes MCP tools for queued sbatch submission, direct srun execution, job monitoring, and cluster inspection. The server supports both scheduler-visible parameter sweeps and lightweight direct execution for debugging.

Requirements

This server requires that Slurm is installed and available on the host system.

Server Configuration

To use the Slurm server, add a slurm entry to your server configuration file.

"slurm": {
    "host": "localhost",
    "port": 8102,
    "transport": "streamable-http"
}

"slurm": {
    "transport": "stdio"
}

Available MCP Tools

Tool Name Description
submit_jobs Submit generated run manifests. Defaults to queued sbatch submission.
check_job_status Check the status of submitted job sets.
continuously_check_job_status Poll job status until a bounded wait condition or timeout.
submit_command Run one ad hoc command using srun; not for generated run manifests.
list_queue List all jobs in the Slurm queue using squeue.
get_cluster_info Get information about the cluster nodes using sinfo.

Tool Selection

Use submit_jobs for generated simulation/orchestration runs. Leave blocking at its default false for normal scheduler-backed workflows because it queues runs with sbatch and returns real Slurm job IDs. Set blocking=true only for lightweight direct debugging when queued jobs are not needed. Use submit_command only for one known ad hoc command, not for run_instances.json or parameter sweeps.

Use check_job_status(job_set_id=...) after submit_jobs(blocking=false) to refresh and report the returned job set plus real Slurm job IDs. Use continuously_check_job_status only when the MCP server should wait for all selected jobs to reach terminal status, or with wait_until="any_running" when downstream work should start as soon as at least one job begins running.

Key Features

  • Queued Job Submission: Submit parameterized runs through sbatch and track real Slurm job IDs.
  • Parallel Direct Execution: Launch multiple parameterized runs directly using srun, with control over concurrency.
  • Single Command Execution: Run arbitrary shell commands on the cluster using srun, with customizable resources.
  • Job Monitoring: Query the status of submitted job sets or individual jobs.
  • Bounded Status Waiting: Poll scheduler status with explicit interval and timeout controls for tests and downstream workflows.
  • Queue Inspection: List all jobs currently in the Slurm queue.
  • Cluster Information: Query cluster node and partition status using sinfo.
  • Configurable Server: Adjust host, port, and transport parameters to fit your cluster environment.