dev-tools 7 min read

MySetup: sync Claude Code and Codex skills across machines

MySetup syncs Claude Code and Codex skills across your machines with a local-first CLI, and doubles as a community for publishing the AI setups people actually use.

By
Share: X in
MySetup social preview card with the mysetup.ai logo and the headline What is the craic with your AI setup?

TL;DR: MySetup keeps the agent skills you write, the SKILL.md files that Claude Code and Codex read, in sync across every machine you use, through a headless CLI that runs on servers and dev containers too. The same site hosts a community where people publish and browse the setups they actually work with. Install takes one Homebrew command or one shell script, and the local library works without creating an account.

What it is

MySetup ships two products on one domain, and they only touch at the edges.

The sync half is called MySetup Sync. It keeps your agent skills in step across machines. A skill is a SKILL.md file, the instruction folder that Claude Code and Codex pick up, so the unit being synced is the thing you actually edit. Edit a review skill on your laptop and the copy on your Linux box catches up. The engine runs headless, which means it works on a desktop, a remote server or a dev container just as well as on a Mac. There is a Mac app if you want a window to browse your library, and a companion iPhone app for quick edits on the move.

The community half is a public directory of setups. Each one is a person’s account of what they use: the tools, the agents, the habits. You can browse every setup without signing in, follow individual people, and subscribe to collections. Signed in, you can publish your own.

These two halves are deliberately separate. Syncing to your own machines never publishes anything. Publishing is a separate action that you review before it goes live.

Why it matters

Skills are the file format nobody standardized. You write one on your laptop, it works on your laptop, and it is missing everywhere else. Fix that by hand and you end up with the same SKILL.md committed in three repos, drifting a little each week. MySetup Sync is one sync engine pointed at that problem, and it keeps the blast radius small: only the skills you pick sync, and the rest of your machine is untouched.

The community side tackles a different gap. Tooling churns fast enough that written-down setups go stale. One commenter on the launch thread put it plainly: “The setups I wrote down three months ago are already half wrong, tools churn that fast.” A directory that shows last-updated dates, change notes and public feeds answers that better than a personal notes file does, because the person who wrote it can amend it when their stack changes.

There is also a group mechanism for teams, where members pick the collections they want and approved updates arrive automatically for their agents.

Getting it running

On macOS, Homebrew handles it and keeps the binary current with your other tools:

brew install mysetupai/tap/mysetup

On Linux the install script pulls the archive for your architecture from mysetup.ai, verifies its minisign signature and its SHA-256 checksum, and drops the binary into ~/.local/bin without needing root:

curl -fsSL https://mysetup.ai/install.sh | sh

You can read that script before running it at mysetup.ai/install.sh. If you would rather verify by hand, download the archive from the download page and check it against the published checksum file:

shasum -a 256 -c --ignore-missing SHA256SUMS.txt

Then walk through the guided setup, which handles sign-in, lets you choose which skills sync privately, and pulls any account skills you already have onto this machine. You can stop at any point and finish later.

mysetup onboard
mysetup doctor

mysetup doctor is the first thing to run when something is not behaving. On a remote machine you reach over SSH, use the device code flow so you can approve from a browser elsewhere:

mysetup onboard --device-code

Enrolling a skill and switching on background sync is two commands. The service keeps running after you log out, so changes propagate without you remembering to push anything.

mysetup sync enroll ./skills code-review
mysetup service enable

Groups use a review-then-subscribe model. You list the group, look at the collections inside, review one, and subscribe when it is approved.

mysetup groups list
mysetup groups collections design-team
mysetup groups review design-team brand-voice
mysetup groups subscribe <review-id>

A few commands worth memorizing:

CommandWhat it does
mysetup library listshows the skills in your local library
mysetup library showprints one skill with its instructions
mysetup sync statusreports the last sync and pending changes
mysetup sync conflictssurfaces cases where both sides changed
mysetup auth statuschecks sign-in and vault state
mysetup self uninstallremoves the CLI cleanly

What to watch for

The CLI is a preview. The docs say plainly that commands may change before the full release, so do not build scripts around them yet.

The engine is closed source. There is no public repository for the MySetup binary itself. What is on GitHub is the distribution plumbing: mysetupai/homebrew-tap for macOS and mysetupai/scoop-bucket for Windows, which its own description calls a preview. The site lists Mac and Linux as the supported platforms, the Mac app needs macOS 14 or newer, and the iPhone app needs iOS 18 or newer.

The MCP pushback is the most interesting part of the HN thread. The share flow suggests connecting an agent over MCP so it can draft your setup for you, and a number of people refused on sight. One wrote that he was not going to connect an arbitrary MCP server and his GitHub account to it, and another said he uses almost no MCP servers and would like a way to share without one. The founder replied that MCP contribution was always optional and that the page has since been changed to make writing a setup by hand the more prominent path. If MCP is not something you want in your setup, the manual editor is now the default route.

On privacy, sign-in runs through GitHub, Google or X. The site states it does not request repository, inbox, messaging or posting access, and it adds no advertising or third-party analytics scripts. View counts use a temporary identifier derived from your IP address that rotates daily, and the raw address is not stored.

Pricing is the soft spot. There is no pricing page, the sync engine is described as free, and the local library works without an account. Groups and the community directory are hosted features that need one. Team features are described on the sync page without a price attached, so treat that as unpriced rather than free.

Finally, the community is young. One commenter, looking for local-AI setups doing real work on limited hardware, noted it still looked like it was waiting on more people to share. The site itself says setups are reported experiences, not verified benchmarks or recommendations.

Source and Accuracy Notes

  • Project page: https://mysetup.ai — verified via direct visit
  • Sync product and plan details: https://mysetup.ai/sync — verified via direct visit
  • CLI reference: https://mysetup.ai/docs/cli — verified via direct visit
  • Sharing, privacy and sign-in details: https://mysetup.ai/sharing — verified via direct visit
  • Distribution repositories: github.com/mysetupai/homebrew-tap and github.com/mysetupai/scoop-bucket — verified via GitHub API
  • License: no source repository and no license file are published for the MySetup binary, and the tap and scoop buckets carry no license file either
  • HN launch thread: https://news.ycombinator.com/item?id=49740105 (251 points, 140 comments, 17 September 2026)