Quickstart

In the next five minutes we'll install Stream Agent Skills and run our first invocation. By the end your agent will be answering Stream SDK questions from live documentation, and you'll know how to reach for the right skill from here on.

You'll need:

An AI coding agent that reads standard skills locations, for example Claude Code, Cursor and Codex.

Step 1. Install the CLI and skills

The getstream CLI is the Stream Dashboard in your terminal. Your coding agent uses it to configure your app, send API requests, and pull data, and its init command also offers to install the skills.

Install the CLI:

curl -fsSL https://getstream.io/cli.sh | bash

Then initialize your project from your working directory:

getstream init

getstream init authenticates with Stream, lets you pick an app (or create one), writes your project credentials, and then offers to install the skills. Pick your coding agent at the prompt (Claude Code, or the universal .agents location that Cursor, Codex and other tools read) and the skills are linked into your project's skills directory (.claude/skills/ for Claude Code). The skills themselves are markdown, nothing about them executes. If you skipped the prompt, run getstream skills at any time.

Step 2. Verify your agent picked them up

Open your agent. You should now see /stream, /stream-docs and /stream-builder available as slash commands:

❯ /stream

/stream          Stream router for Chat, Video, Feeds, and Moderation. Use when
                 the user wants to build a new app with Stream, scaffold a…
/stream-docs     Search live Stream SDK documentation for Chat, Video, Feeds,
                 and Moderation. Look up how a Stream React/iOS/Android hook…
/stream-builder  The framework-agnostic Stream builder (Steps 0-7 scaffold +
                 enhance + Video audit), being extended to support app kinds…

The web pack (/stream-react) and the platform packs (/stream-swift, /stream-android, /stream-react-native, /stream-flutter) aren't installed yet. The router adds the one you need the first time a task calls for it.

If they aren't there, restart your agent so it rescans its skills directory.

Step 3. Your first invocation

The easiest skill to try first is /stream-docs, because it doesn't need credentials or any other setup. It looks up live SDK documentation and answers with a citation.

Type this into your agent:

/stream-docs how do I generate a user token in Node.js?

You should get a short answer that includes a code snippet using serverClient.createToken(), with a citation pointing at a getstream.io documentation URL:

❯ /stream-docs how do I generate a user token in Node.js?

Generate the token on your server with the API secret, then hand it to the
client. Never create user tokens in the browser.

  import { StreamChat } from "stream-chat";

  const serverClient = StreamChat.getInstance(apiKey, apiSecret);
  const token = serverClient.createToken("jdoe");

Source › https://getstream.io/chat/docs/node/tokens-and-authentication/

Click the citation. The code in the answer matches the code on that page: every answer is pulled from the current docs with the source attached, so you can always check it.

Step 4. Let the router pick the skill

When you don't know which skill applies, invoke /stream and let it dispatch. Try:

/stream show me how to scaffold a Next.js chat app

The router reads the prompt and recognises it as a web scaffolding request for /stream-react, the default web pack. That pack isn't part of the default install, so the router installs it on demand first, then hands off. You'll see /stream-react take over and start asking about your project:

❯ /stream show me how to scaffold a Next.js chat app

stream: classifying request… scaffold a new web app
stream: /stream-react isn't installed yet - running `getstream skills stream-react`
stream: dispatching to /stream-react

stream-react: I'll scaffold a Next.js chat app. A couple of questions first:
  1. What should I name the project?
  2. Add Video calling alongside Chat? (y/N)

You can cancel at the first prompt if you don't want to actually scaffold. The point is to see the routing work.

What you've done

In four steps you've:

Installed the getstream CLI and the core Stream skill pack
Confirmed your agent picked the skills up
Used /stream-docs to pull a live, cited SDK answer
Seen /stream install a missing pack on demand and route a vague prompt to it

Where to go next

To scaffold a real web app: read the /stream-react reference for what it can produce and the prompts it understands.
To query your live Stream data: the CLI is built into the router. If you ran getstream init in step 1 you're already set. Just ask /stream questions like "list my channels" in plain English.
To build for a native platform: pick the right platform skill. /stream-swift, /stream-android, /stream-react-native or /stream-flutter.
To understand the trust model: Security and trust covers everything that can touch the network and when.

Welcome

Install