Claude Science knows how to connect via SSH to execute work on compute servers and HPC clusters
That is good, but not very good. It’s 2026, and who wants to hang an SSH server out on the naked, public Internet?
What I really need for Claude Science and my compute …
This should be table-stakes in 2026:
- I want to run my AWS infrastructure in private subnets; no internet exposure at all
- I don’t want to have to use or connect to an internet-facing SSH server
Confirmed! We can proxy AWS SSM traffic over the Claude Science SSH transport layer
That’s the post folks. I was able to get hands-on today with Claude Science and wired it up to a few different private HPC clusters running in AWS — with all connectivity being securely handled by secured AWS SSM sessions proxied over the Claude Science SSH layer.
Since the goal was to test, document and share my methods it was better to place all the material in a github repo rather than here so you can see the current state over at this Github Repo
https://github.com/bioteam/claude-science-hpc-integrations/tree/main
If you want to know how to create a least-privileged IAM user that can only make SSM calls to a single host, start here:
– https://github.com/bioteam/claude-science-hpc-integrations/tree/main/iam-user-for-ssm-sessions
If you already understand SSM and just want to wire up Claude Science to a generic AWS HPC cluster running Slurm start here:
If you are a computational chemist and want to wire Claude Science up to an HPC cluster running Schrodinger Software than start here:
Hope this helps!
