Quickstart: Connect Hyper to Your AI Client

Getting Hyper running takes about five minutes. You’ll install the macOS desktop app, authenticate with Google, and configure your AI client — Claude, Cursor, or Codex — to connect to Hyper’s MCP server. After that, every session your agents run will automatically brief itself from your team’s shared memory.

Hyper is currently macOS-only. Windows support is on the roadmap — sign up at heyhyper.ai to be notified when it ships.

Prerequisites

Before you start, make sure you have:

A Mac running macOS 12 Monterey or later
At least one AI coding client installed: Claude Code, Claude Desktop, Cursor, or OpenAI Codex CLI
A Google account for authentication

Setup steps

Download and install the desktop app

Visit heyhyper.ai and click Download for macOS. Open the downloaded .dmg, drag Hyper into your Applications folder, and launch it.The desktop app is built on Tauri and runs natively on macOS — no Electron overhead, no browser tab to keep alive.

Hyper uses Google OAuth for authentication. Click Sign in with Google in the app and complete the OAuth flow in the browser window that opens. Once authenticated, you’ll land on the workspace selection screen.

Your Google credentials are never stored by Hyper directly — only the OAuth token issued by Google is used to authenticate your session.

Create or join a workspace

A workspace is the shared memory space for your team. You have two options:

Create a new workspace
Join an existing workspace

Click New Workspace, give it a name (usually your company or project name), and choose whether it’s a Personal or Team workspace. You’ll become the admin automatically.

Not sure which to pick? Start with a personal workspace to try Hyper solo, then create or join a team workspace when you’re ready to share memory with colleagues.

Connect your AI client

Hyper exposes an MCP server your AI client connects to. Choose your client below and follow the configuration steps.

Claude Code / Claude Desktop
Cursor
Codex CLI

Open your Claude configuration file (~/.claude/claude_desktop_config.json for Claude Desktop, or the equivalent for Claude Code) and add the following entry under mcpServers:

{
  "mcpServers": {
    "hyper": {
      "url": "https://hyperlink.gethyper.space/mcp"
    }
  }
}

Restart Claude after saving.

In Cursor, open Settings → MCP Servers and add a new server:

Name: hyper
URL: https://hyperlink.gethyper.space/mcp

Save and restart Cursor.

Add the MCP server to your Codex configuration file:

{
  "mcp": {
    "servers": {
      "hyper": {
        "url": "https://hyperlink.gethyper.space/mcp"
      }
    }
  }
}

Verify with the connect tool

Start a new session in your AI client and ask it to call the connect tool. Hyper will authenticate the session, register the agent, and return a briefing — a digest of your team’s current context pulled from your workspace memory.A typical connect call looks like this:

connect()

Or, if you’re joining with a specific workspace token and timezone:

connect(join_token="abc123", tz="America/New_York")

Example briefing response:

✅ Connected to workspace: Acme Engineering
📋 Briefing as of 2025-06-10:

Current sprint goal: Ship the billing migration by June 20.
Auth stack: Supabase (RLS enabled for multi-tenant isolation).
API versioning decision: Prefix all new routes with /v2/.
Alice is leading the payments module; blocked on Stripe webhook retry logic.

If you see a briefing, Hyper is working correctly.

Run your first ask

Now test memory retrieval directly. Ask your AI client to call the ask tool with a question about your project:

ask("What auth library are we using and why?")

Hyper searches your workspace memory — including past decisions, team notes, and synthesized context — and returns a grounded answer:

We're using Supabase Auth. The decision was made on 2025-05-14 because it
includes Row-Level Security (RLS) out of the box, which is required for
multi-tenant isolation. Firebase Auth and a custom JWT approach were
evaluated but ruled out due to complexity and maintenance overhead.

If the answer comes back empty, it just means no one has recorded that memory yet. Use remember() to capture important decisions as you make them.

What’s next

You’re connected. Here’s where to go from here:

Claude Setup Guide

Detailed Claude Code and Claude Desktop configuration

Cursor Setup Guide

Step-by-step Cursor MCP configuration

How Memory Works

Learn how Hyper classifies and stores what agents remember

Invite Your Team

Add teammates and share a workspace

Get Started

Core Concepts

Integrations

Guides

Plans & Billing

FAQ

Quickstart: Connect Hyper to Your AI Client

Prerequisites

Setup steps

What’s next

Claude Setup Guide

Cursor Setup Guide

How Memory Works

Invite Your Team

​Prerequisites

​Setup steps

​What’s next

Claude Setup Guide

Cursor Setup Guide

How Memory Works

Invite Your Team

Prerequisites

Setup steps

What’s next