Quickstart - from an app your agent built to coworkers using it

This page takes you from a finished app to colleagues using it. You need an app (or an agent that can build one), Node.js 22 or newer on your computer, and one of the supported agents: Claude, Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI or Antigravity.

1. Connect your AI once #

Open a terminal and run:

npx -y scava setup

It adds Scava to the agents it finds on your computer. See Connect your agent for what is added where. If the Codex app is open, restart it so it loads Scava.

For Claude on the web, in the app, on your phone, and in Claude Code when you are signed in with Claude: in Claude, open Customize, then Connectors, choose "Add custom connector" and enter the address https://scava.ai/mcp. The connection then lives in your Claude account.

The first time your AI needs Scava it opens a page in your browser. Sign in to Scava with Google or an emailed code, check what the page says, and choose Allow. No key stays on your computer. You can see and disconnect every connected agent on the Connect page at scava.ai.

2. Get an app #

Ask your agent to build the tool, the way you normally would. Two things are specific to Scava:

The app is a folder with an index.html at its top. A build step is fine: publish the output folder (for example dist) and use relative paths. With Vite set base: './'.
Do not build a login and never add password fields. Scava signs people in, and an upload with a password field is refused.

To store data that all your colleagues see, the app calls scava.db('name'). Your agent learns this when it connects, because Scava hands it a short reference. If you want to see what your agent reads, run npx -y scava docs.

A good first request to the agent:

Build a vacation planner for my team. Everyone enters their own vacation,
everyone sees everyone's entries. Store the data with scava.db.

3. Publish it #

Tell the agent:

Publish this folder to Scava.

The agent calls the publish tool and runs the one-time command it gets back; in a chat it passes the files themselves. Scava checks the upload, creates version 1 and returns the address, for example https://team-vacation-k3x9.scava.site. The app is private: only you can open it.

Right after publishing from a folder, the command opens the live app once in a headless Chrome or Edge on your computer and lists what went wrong: script errors, calls the browser blocked, missing files, refused data calls. Each problem comes with a hint. Your agent reads it and fixes the app, then publishes again. See Host an internal tool for the test run.

Open the address yourself and use the app. You see it exactly as your colleagues will, including the "Suggest a change" button in the corner.

Tell the agent who may open the app:

Share it with @mycompany.com.

You can share with single email addresses, with a whole company domain, or with a group. Single addresses get an invitation email. See Sharing.

Your colleagues open the link, choose "Continue with Google" or "Send me a code", and are in. A person who is not on the list sees a "No access" page with a button to request access; you get an email and approve with one click.

5. Let colleagues ask for changes #

Each person who uses the app can click "Suggest a change", describe a wish and point at the spot on the page. You get an email. If you accept the request, your agent can fetch it:

Get my accepted change requests and build them.

The agent builds the change and publishes with the request numbers. The person who asked receives an email that the change is live. Scava never changes your app on its own. See Change requests.

Next steps #

Decide who may read and change what: Permission rules
Store uploads such as photos and PDFs: Files
Call an AI or a company system from the app: Server code
Use an address such as tools.company.com: Your own address