A machine joins your workspace by running myco — a single self-contained binary with no Node, Python or Docker required. Once its daemon is up, that machine shows up in Mycolab: you can spawn agent sessions on it, browse and transfer its files, expose its local ports to public URLs, and run scheduled work — from a browser, anywhere.
About a minute on a fresh machine, and nothing to copy between them. Both steps run on the new machine; you approve it in a browser — any browser, including the one on your phone.
On the new machine (macOS or Linux, Intel or ARM):
curl -fsSL https://hypha.run/install.sh | sh
Installs to /usr/local/bin/myco (uses sudo only if that directory is not writable). No root needed if you set PREFIX=$HOME/.local/bin. The installer will not silently overwrite an existing myco — it asks, or refuses when piped; pass FORCE=1 to replace it or NAME=myco2 to install alongside. Prefer a direct download? See the table below.
On the same machine:
myco login myco daemon start myco daemon status # should report the daemon running
myco login prints a URL, a short code and a QR code, and opens your browser if it can. Approve it, and the machine picks up its own credential within a few seconds. There is nothing to copy back.
Over SSH, or on a box with no browser? The same command works: scan the QR with your phone, or open the printed URL on any device you are logged in on and type the short code. Add --no-browser to skip the launch attempt. Credentials are written to ~/.mycolab/credentials.json.
# the whole thing, start to finish curl -fsSL https://hypha.run/install.sh | sh myco login myco daemon install # run at boot (launchd / systemd), instead of 'daemon start'
It generates a long random code and never shows it. Approving in the browser records your consent and returns no token; the credential goes to whoever holds that code, which is the machine. The short code you type does nothing on its own.
It rotates itself, so the disk never holds a long-lived admin key. Revoke a machine from Machines — by name, not by token id.
No browser, nobody to click. Mint a scoped token from CI & automation tokens and pass it as myco login --token ….
A server you walk away from needs two things the quick-start skips.
myco daemon install registers the daemon as a system service — launchd on macOS, systemd on Linux — so it comes back after a restart. myco daemon uninstall removes it. Plain daemon start does not survive a reboot.
A machine connected with myco login refreshes its own credential, so the old failure — the token quietly expires, the machine stops appearing, and the only sign is a log line — does not happen. A machine still using a pasted --token does expire on schedule: if one goes quiet with no other explanation, suspect the token, and re-run myco login to move it off the paste entirely.
myco daemon status on the box, or just look for the machine in Mycolab — presence there is authoritative, derived from the live connection rather than self-reported.
Two commands cover almost everything: one for a directory of files, one for a port something is already listening on.
Turns a local server into a public https://tl-*.hypha.run URL over the mesh's built-in tunnel — WebSockets included, no firewall changes, no port forwarding. It uses the credential the machine picked up at login; there is nothing else to sign in to. One service, one port.
Publishes a directory — a build, a report, a static site — at its own hostname, so an app's absolute paths work with no rebuild. Private by default: the URL is an unguessable hostname, and anyone holding it can read it. Use --owner or --access a@x,b@y when that is not what you want, or --public when it is.
An error mentioning a different login command, or a frontend host you never configured, means a different binary of the same shape is earlier on your PATH. Check with which -a myco and myco --version. myco is a single static binary — no Node, no Python, no separate login step beyond myco login.
Statically linked, dependency-free. The Linux builds are musl-static and run on any distribution — Debian, Ubuntu, Alpine, or a distroless container alike.
Unpack and put it on your PATH: tar xzf myco-<target>.tar.gz && sudo mv myco /usr/local/bin/. Windows is not supported — the daemon is Unix-only; use WSL2, where the Linux x86_64 build works.
Everything below is driven from the browser, against the machine you just connected.
myco session spawn claude -d ./project — or spawn one from Mycolab. The agent runs on your machine, with your files and your tools; the browser is only the window onto it.
Read, upload and download from the machine's disk over a streaming tunnel — not a control channel — so large files work without buffering.
myco expose 3000 gives a dev server on your laptop a public HTTPS URL, WebSockets included. No firewall changes, no port forwarding.
myco serve report ./out publishes a directory — private to your account by default, public on request.
Declare long-running processes in YAML with keep-alive and health probes: myco process apply proc.yaml.
Cron lives server-side, so a schedule still fires while the machine is briefly down, and runs when it comes back.
myco attach <workspace> adds a second workspace beside the one you just logged into — its own credential, its own connection, the same machine in both lists. Add --no-agents to give that workspace a shell here and no agent sessions; the role is per workspace. myco workspaces shows what this computer serves.
Open Mycolab to see the machine, spawn an agent session on it, browse its files, and drive everything from the browser.
Run myco --help for the full command surface. Everything on this page is the shipped CLI — no preview flags.