william-dawson

One MCP server for every onboarded HPC facility — submit jobs, monitor queues, manage files and search per-machine guides across nine supercomputers through one facility-parametrized tool surface.

Shared runtime for the HPC MCP agent family: SSH middleware, PSI/J-style job models, scheduler backends, docs RAG, doctor, and serving, reused across per-machine plugins (Rikyu, HOKUSAI, Octopus, ...).