How to put an app from a GitHub repo in front of your team
The code is in a GitHub repository and a team needs to use the app, behind sign-in. What GitHub Pages, Codespaces and general hosts each do well, what a repository needs before it publishes cleanly, and the one command that publishes it with Spryloom.
A lot of internal tools now start as a repository. An engineer asks an agent for a deploy-freeze calendar and pushes it. Someone in finance ops has a colleague put the expense checker on GitHub "so it's safe". A support lead finds a starter on GitHub that's close to the triage board they want.
The code is in a good place. The team still can't use it. A repository is somewhere code lives, not somewhere an app runs, and the people who need the tool shouldn't need a GitHub account to open it.
This article is about getting from the repository to a running app your team signs in to. We make Spryloom, and the last option is ours, so read that section knowing who wrote it. The other options are described as fairly as we can.
What "in front of your team" needs
- Somewhere it runs, including its database if it keeps anything, while your laptop is closed.
- A door: the team signs in, and nobody else gets in, including someone who was forwarded the link.
- A repeatable build: the same code gives the same app, every time you publish.
- Secrets kept out of the repository: API keys set on the running app, not committed.
Option 1: GitHub Pages
GitHub Pages serves a static site straight from a repository, and it's free and simple. For docs, a project page, or a front end with no server, it's a good answer.
It serves files only, so an app with a server or a database won't run there. And sites are generally open to anyone with the address. Restricting a Pages site to certain people is limited to particular GitHub plans, and viewers need GitHub accounts. Check your plan's current terms if that's the route you want.
Option 2: Codespaces
A Codespace runs the repository in a cloud machine, and you can forward a port so someone else can open the running app. It's a good way for a colleague to try a branch or review a change.
It's a development environment, not hosting: it stops when idle, belongs to one person's account, and isn't meant to keep a team's tool running every day.
Option 3: a general host
Vercel, Render, Railway, Fly and others connect to a GitHub repository and deploy it, often on every push. They run nearly anything, and developers know them well. For a public product, one of these is the usual choice.
For an internal tool, the parts that remain are yours: sign-in for your team, a database and its connection string, and the deployment settings. An engineer can do all of that. A team without one usually stops here.
Option 4: publish the repository with Spryloom
Spryloom runs apps for teams behind sign-in. The app contains no login code: Spryloom checks who each visitor is before a request reaches the app, and tells the app their address. Each app gets its own small machine and, if it asks for one, its own Postgres database.
From a terminal, the repository's address is enough. A branch, tag or commit goes after an @:
npx -y spryloom publish github.com/you/expense-checker # a branch, tag or commit spry publish github.com/you/expense-checker@release # one app in a repository with several spry publish github.com/you/tools/tree/main/apps/expense-checker # a zip of the folder spry publish ./expense-checker.zip
A zip of the folder works the same way. What goes up leaves out these, and Spryloom makes the build output again when it builds:
A private repository
Set GITHUB_TOKEN in your terminal before you publish. It's sent to GitHub to fetch the code and reaches Spryloom at no point.
A repository with several projects
Publish the app's own folder, not the root: copy the folder's address from GitHub's address bar (the third command above), or write it as github.com/you/tools/apps/expense-checker@main. A folder with no package.json and no index.html is refused, because guessing which subfolder you meant would be worse than asking.
What the repository should hold
Most first publishes that fail were decided in the repository. Four things make it work the first time, and the Spryloom skill teaches them to your agent while it builds:
- A committed lockfile. Without
package-lock.json, the build installs whatever is newest that day, not what you tested. - A start script, and the port from the environment.
"start": "node server.js", and a server onprocess.env.PORTat0.0.0.0. spryloom.yaml, committed. What the app is, who may use it, what it needs. Every publish from any machine gets the same answer.- No secrets. Not in the code, not in
spryloom.yaml, not in a committed.env.
Set secrets on the app instead:
spry secrets expense-checker set STRIPE_API_KEY=...
A secret reaches the app when it's published, so publish again after setting one. Nobody can read a secret back, including you.
A minimal manifest for the expense checker:
app: name: Expense checker description: Flags expense claims that need a second look, for the finance team. runtime: backend: node start: node server.js access: visibility: company domain: yourcompany.com data: postgres: true tables: [claims, flags]
- What it is, in one sentence for the team.
- A Node server, started by its start script.
- Anyone who signs in at yourcompany.com. Needs a company workspace.
- Its own Postgres database, with these tables.
With visibility: invited, you name the people instead:
spry invite expense-checker --email emily@yourcompany.com
spry check in a local copy shows what Spryloom will see, and whether it can publish, without publishing anything.
After it's live
Your team opens the address, signs in with an emailed link, and uses the app. No GitHub account, no password. The publish gives the address, for example https://expense-checker.yourcompany-com.spryloom.app. Publishing again keeps the address, the data and the people, and raises the version. spry rollback expense-checker returns to the previous one.
What it doesn't do
- It doesn't watch the repository. A push doesn't publish. Run
spry publishagain, or ask your agent, when you want a new version live. - It builds Node. One language and one runtime: a Node server, with a static or React front end. Spryloom doesn't build a Python or Go app.
- The app reaches almost nothing outside. It has its own database. Other hosts are blocked unless declared from a fixed list, such as
api.openai.com,api.github.comorapi.stripe.com. An app that calls an arbitrary API needs that host added first. Ask us at hello@spryloom.com. - It's for people who sign in. A public website belongs on a general host.
- No file uploads. Spryloom has no file storage for an app, and
uploads: trueis refused. Keep what matters in the database. - Limits. A workspace holds 10 apps, each with up to 25 invited people and 20 secrets.
If the repository is a front end with no server, it becomes a page instead of an app. How to publish a Vite or React app with no backend, behind a login covers that case.
Start here
From a terminal:
npx -y spryloom publish github.com/you/app
In Claude Code, add the plugin once, then ask it to publish:
/plugin marketplace add spryloom/claude-plugin
/plugin install spryloom@spryloom
In Codex:
codex mcp add spryloom -- npx -y spryloom mcp