ParallelClusterMaker is now available as an MCP server, allowing users to provision and manage AWS ParallelCluster stacks using Claude Code and https://claude.ai in the browser along with the CLI. The necessary infrastructure deploys into your specified AWS account and can be heavily customized to support a wide range of HPC and ML training use cases. Furthermore, cluster stacks are universally visible regardless if they were created with the CLI, Claude Code, or https://claude.ai.
Common operations like building/destroying cluster stacks, modifying compute/gpu queue configurations, and starting/stopping compute fleets are all supported. Please note that submitting jobs is available only from Claude Code or the CLI, but you can perform all other read-only operations like inspecting job statuses, reviewing queue configurations, and viewing cluster/node states in Claude.ai. Additionally, Slurm accounting with a dynamically configured lightweight local MariaDB backend is now enabled by default, providing Claude with the ability to help troubleshoot cluster jobs and configuration issues.
You might find this useful for managing AWS ParallelCluster stacks without having to configure that YAML file on your own or babysit the AWS console. If things fail, I’ve included mechanisms to help with troubleshooting so you don’t have to waste time digging through CloudWatch or instance logs. Good luck, hope it helps with your model training, protein docking, or genomic data analysis tasks!
reply