How to publish an app you built with Codex for your team
Codex built the tool and it runs on your machine. How to get it to your team behind sign-in from the same Codex session, what Codex writes along the way, the two questions it asks, and what your coworkers see.
An engineering manager asks Codex for a small tool: a page where the team records which on-call shifts they have swapped, so the rota stops living in a chat thread. Codex writes a Node server, a few screens and a table of swaps. It runs at localhost:3000, and it works.
Now the six people on the rota need to use it. They shouldn't need to install anything, and nobody outside the team should be able to open it.
This article is about getting a Codex-built app from your machine to those six people. We make Spryloom, and most of what follows is how it works with Codex, so read it knowing who wrote it. If someone on your team would rather run the app on a general host, How to share a Claude Code app with your team describes those routes fairly, and they apply to Codex apps the same way.
The short version
- Add Spryloom to Codex, once.
- Ask Codex to publish the app for the people who should use it.
- Answer the two questions it asks.
- Invite people.
1. Add Spryloom to Codex
In a terminal:
codex mcp add spryloom -- npx -y spryloom mcp
codex mcp list should now show spryloom. Restart Codex if it was open, since it reads its servers when it starts. Nothing needs installing first: npx fetches the server the first time. You need Node 20.19 or newer, which you already have if Codex is running Node projects.
The first time Codex uses a Spryloom tool, it says Spryloom isn't signed in and asks for your email. Spryloom emails you a link, Codex shows a short code, and you open the link and check the page shows the same code. Then tell Codex you've confirmed, and it carries on. This happens once per machine, and the spry command shares the same sign-in.
The guidance Codex works from
Most failed publishes are decided while the app is being written, not when it is published. The rules are short:
- Read the port from
PORTand bind to0.0.0.0. - Commit a lockfile and have a
startscript. - Keep what must last in Postgres, not on disk.
- Read settings and keys from the environment.
Spryloom writes them down as a skill, and Codex reads skills. In the app's folder:
npx -y spryloom skill
mkdir -p .agents/skills
mv .claude/skills/spryloom .agents/skills/
spry skill writes the skill into .claude/skills/spryloom/, where Claude Code looks. Codex looks in .agents/skills/, so the move puts it where Codex finds it.
Without the skill, Codex still publishes. The Spryloom server checks the folder first: an app with no way to start is refused with what to fix, and a fixed port, a missing lockfile or a committed .env file is listed as worth fixing.
2. Ask Codex to publish
Ask in the session where Codex built the app:
Publish this with Spryloom for the on-call team. Only people I invite.
Before publishing, Codex writes spryloom.yaml in the app's folder, as the skill tells it to. It knows what the app needs because it wrote it. (Without the skill, the publish tool writes one from what Codex passes it and what it finds in the folder.)
app: name: On-call swaps description: Record on-call shift swaps so everyone sees the current rota. runtime: frontend: static backend: node start: node server.js access: visibility: invited admins: [jordan@yourcompany.com] data: postgres: true tables: [swaps, people] uploads: false
- The description appears in the invitation and on the label page. One plain sentence.
- A Node server with plain-file screens.
- Sign-in for invited people. Admin rights for
admins; the publisher is always one. - A database, because
postgresis true.
Spryloom creates exactly what the file declares. The description is how a coworker decides whether to open something a colleague made with AI. The file holds no secrets; it is committed to the repository.
3. The two questions
With the skill, Codex asks two things, and works out the rest. Without it, the publish tool still tells Codex to ask the first: who should be able to use it?
Just you
The default, and right while the app is half-finished.
People you invite
The on-call team.
Anyone at your company's domain
A tool the whole company uses. Needs you signed in with a work address.
The second question comes only if the app stores data: Codex confirms the table names it is about to declare, because coworkers see them on the label page. "swaps and people" is something a teammate can read and understand.
Then the publish runs. It takes a few minutes for an app: Spryloom builds it in its own sandbox, sets up its database, and waits for it to answer before switching over. Codex can tell you which stage it is on. When it finishes, it reports the address, who can sign in and what was created, along these lines:
Published: https://on-call-swaps.yourcompany-com.spryloom.app
Sign-in: people you invite
Data: 2 tables
Want me to send this to anyone?
If a publish fails, the message says what is wrong. When the cause is one of the rules above, such as a fixed port, it is Codex's mistake to fix, and it should fix it and publish again.
4. Invite people
Ask Codex ("invite alex@yourcompany.com and taylor@yourcompany.com"), or from a terminal:
spry invite on-call-swaps --email alex@yourcompany.com
Alex gets this:
Subject you@yourcompany.com invited you to On-call swaps
You have been invited to On-call swaps
you@yourcompany.com invited you to use it.
What it does: Record on-call shift swaps so everyone sees the current rota.
Open On-call swapsThis opens on-call-swaps.yourcompany-com.spryloom.app
It signs you in as alex@yourcompany.com. It works once, for 7 days; after that, sign in with an emailed link.
No account to create, no password, nothing to install.
Inside the app, Alex is Alex. Every request reaches the app with x-spryloom-email and x-spryloom-role headers, so "swapped by Alex" comes from the sign-in, not from a name typed into a form. Spryloom strips any copy of those headers sent from outside, so the app can trust them.
Someone who wasn't invited reaches the sign-in page and gets nothing, and your Spryloom dashboard lists the address that tried, with an Invite button if you meant to let them in. To remove someone, spry uninvite on-call-swaps --email alex@yourcompany.com works on their next click.
After that
Changes are asking Codex again, in the same folder. The address, the data and the people stay. A few commands are worth knowing:
spry apps every app in your workspace
spry logs on-call-swaps recent output
spry rollback on-call-swaps back one version
spry secrets on-call-swaps set SLACK_BOT_TOKEN=...
The last one is for when you ask Codex to post each swap to the team's Slack channel. The app then declares slack.com in egress, and posts through Slack's bot API with that token. Secrets reach the app when it is published, so publish again after setting one.
What Spryloom doesn't do
- It is for people you know. Everyone signs in. A public site or a product belongs on a general host.
- Sign-in is an emailed link. No passwords, and no Google or Microsoft sign-in.
- Outside services come from a fixed list: Stripe, Slack, GitHub, Linear, Notion, Airtable, HubSpot, Google Sheets, and OpenAI, Anthropic and Gemini. Ask us at hello@spryloom.com for another.
- No file uploads, and the filesystem is read-only apart from a temporary folder.
- Node is the only server language. The app's server must be Node, and its front end can be React or plain files.
- Limits. 10 apps per workspace, 25 invited people per app, 20 secrets per app.
Start here
In Codex:
codex mcp add spryloom -- npx -y spryloom mcp
Then ask it to publish the app for the people who should use it. Without Codex, from the app's folder:
npx -y spryloom publish
Before you send the link, open it in a private browser window. You should see a sign-in page, not the app.