Why mx
You're never working on just one thing. A feature is in flight, an urgent bug lands, a review comes back — and each one wants a different branch, a different set of running services, and a different mental context.
With a single working copy of each repo, only one of those can be active at a time. Switching between them means the same tax, over and over:
git stashwhatever you had in progress- check out the other branch (and create it if it's new)
- reinstall dependencies that drifted between branches
- restart dev servers — and fight "port already in use" when two apps want
:3000 - rebuild the context in your head of where you were
And there's a newer constraint: a coding agent can only work one checkout at a time. If you want several agents making progress in parallel, a single working copy makes that impossible.
"Can't I just use a tab per feature, or subagents?"
You can reach for git worktree by hand, or open a terminal tab per feature, or run several subagents in one session. It works — for about a day. Then the cracks show:
- Ports. Two dev servers, one port number. You start hand-assigning ports and tracking them in your head.
- Hydration. A raw
git worktreeis an empty directory — no.env, nonode_modules, no seeded database. You set it up by hand, every single time. - Multiple repos. A single feature often spans two or three repos (say a frontend and a backend). A plain session lives in one checkout, so the other repos are out of reach.
On top of that: no lifecycle to archive and later recover a feature, no memory shared between features, and a growing pile of tabs, folders, and branches you track by hand.
What mx does
mx is a manager for parallel features. It's built on the primitives you already have — git worktrees and coding agents — and adds the management layer around them:
- Works. One command creates a work for a feature: its worktrees (one per repo it touches), branches, and editor workspace.
- Allocated ports. Every service gets a port that's unique across the whole runtime, so the same app can run many times at once.
- Hydrate hooks. A fresh worktree is made ready to run automatically — copy env, install deps, seed data — via a lifecycle hook you control.
- Context registry. A runtime-wide store of findings, conventions, and runbooks that every feature's agent reads from.
- Lifecycle. Archive a finished feature to free its ports and worktrees while keeping the branches and notes; unarchive to pick it right back up.
- Mission control. A live, read-only dashboard of every work, its ports, and its health — the whole fleet at a glance.
The result: instead of one feature at a time, you run a fleet of features in parallel, each fully isolated, each with its own agent — and switching between them is instant.
Ready to try it?
Head to Getting started to install mx and create your first parallel feature.