@alexrivera.ai

The AI
Arsenal

The exact stack I use to run my life, work & agency.

You commented "AGENCY" — so here it is. Every tool, every real setup step, no fluff. Most of this runs on owned infrastructure and my Claude Max subscription. The paid pieces are listed honestly. Read top to bottom or jump straight to the category you need.

Growth Build Ops Intel

START HERE

3 things to set up first

Everything else in this arsenal plugs into these three. Set them up before anything else.

  1. 01
    Install the Claude Code CLI

    This is the command-line interface that unlocks agents, MCP connections, sub-agents, and custom commands. Install it with npm:

    npm install -g @anthropic-ai/claude-code

    Then run claude in any terminal to authenticate with your Anthropic account. Full docs: docs.anthropic.com/claude-code

  2. 02
    Connect your MCP servers

    MCP (Model Context Protocol) servers let Claude reach into your real tools — Gmail, Calendar, Drive, Notion, Vercel, and more. Each server is a small process that Claude talks to.

    Inside Claude Code, run claude mcp add and follow the prompts, or edit your ~/.claude/config.json to add server definitions. The official MCP registry and setup guide: modelcontextprotocol.io

    Alex runs 11 connected MCP servers. The most-used: Gmail, Google Calendar, Google Drive, Notion, and Vercel.

  3. 03
    Pick your CRM hub (Alex uses GoHighLevel)

    Before layering AI on top of sales and ops, you need one source of truth for contacts, deals, and follow-ups. Alex built his on GoHighLevel (and later Supabase for custom dashboards). Pick one and commit — the agents read from it.

    GoHighLevel: gohighlevel.com

Growth

Growth

The tools that find leads, run ads, create content, and fill the calendar.

Growth

Meta MCP

Run Meta ads from a chat box.

An MCP server that connects Claude directly to the Meta Ads API. Brief a campaign in plain English, pull ROAS and KPIs back into Claude — no Ads Manager rabbit-holes.

WHAT YOU NEED

A Meta Business account, an active ad account, and the Claude Code CLI running with MCP support enabled.

SETUP

  1. Install the Claude Code CLI (see Start Here above).
  2. Find or install the Meta Ads MCP server for Claude Code — search the MCP registry at modelcontextprotocol.io.
  3. Generate a Meta System User token in Business Manager with ads_read and ads_management permissions.
  4. Add the server to your Claude MCP config with the token as an env variable.
  5. Test: ask Claude "What's the ROAS on my current campaign?" — it should pull live data.
Meta Marketing API docs →
Growth

Higgsfield

Ad creatives in minutes, not days.

An AI video and image generation platform built for marketing creatives. Generates studio-quality output — scroll-stopping videos and images — without a videographer or editor. Pairs with the Meta MCP: Higgsfield makes the ad, the MCP runs it.

WHAT YOU NEED

A Higgsfield account and a clear creative brief.

SETUP

  1. Sign up at higgsfield.ai.
  2. Generate ad creatives from a text or image prompt.
  3. Create a project, choose your output type (image or video), write your brief.
  4. Iterate on the output using the in-app editor, then export.
  5. Hand the file to your ad platform or plug into the Meta MCP workflow.
higgsfield.ai →
Growth

Boosend

Every comment becomes a booked call.

An Instagram DM automation tool. When someone comments a keyword on your post — like "AGENCY" — Boosend fires an instant DM with your offer or lead magnet. AI agents qualify and route conversations 24/7. This is the funnel that delivered you this very page.

WHAT YOU NEED

An Instagram Business account, a Facebook Page connected to it, and a Boosend account.

SETUP

  1. Sign up at boosend.com.
  2. Connect your Instagram Business account via the Facebook/Meta integration.
  3. Create a new "Comment Trigger" — set your keyword (e.g., AGENCY).
  4. Write the DM sequence: immediate reply → follow-up → optional booking link.
  5. Set the trigger live on your post before publishing.
boosend.com →
SEO
AEO
Growth

SEO + AEO Automation

Rank on Google and the AI answers — continuously.

Alex's in-house content engine that auto-optimizes his site for both classic Google SEO and Answer Engine Optimization (getting cited by ChatGPT, Perplexity, and other AI tools). It runs continuously — no monthly retainer, no agency.

THIS IS A PROPRIETARY SYSTEM

This is not a public install. Here's the pattern to build your own:

  • Build a city × service content matrix (e.g., 70 pages targeting "pest control [city]") using Claude Code agents to generate and publish pages.
  • Add schema markup (LocalBusiness, FAQ, HowTo) on every page for AEO.
  • Run a daily agent that monitors rankings, identifies gaps, and drafts new content.
  • Stack on a Next.js site deployed to Vercel for fast, SEO-friendly delivery.
Deploy on Vercel →
Build

Build

How the sites, automations, and products actually get made.

Build

21st.dev

Premium UI components, shipped fast.

A library of production-grade, drop-in UI components and an AI generator for building them. Alex starts every site here instead of from a blank file — cutting a polished landing page from weeks to hours.

WHAT YOU NEED

A 21st.dev account and a React/Next.js project to paste components into.

SETUP

  1. Sign up at 21st.dev.
  2. Browse or search for the component you need (hero, pricing, nav, etc.).
  3. Copy the component code and paste into your project, or use the AI generator to describe what you want.
  4. Install any peer dependencies it lists (typically Tailwind + shadcn/ui).
  5. Customize colors and copy to match your brand.
21st.dev →
AUTO
CONTENT
Build

Autonomous Content Engine

A content team that built this carousel.

Alex's in-house system that researches trends, designs and writes Instagram carousels 24/7, and delivers finished posts for a single approval tap. This exact carousel came out of it. It runs on Claude Code agents on his always-on Mac Mini.

THIS IS A PROPRIETARY SYSTEM

Not a public install. Here's the pattern:

  • Research agent: scrapes trending topics in your niche daily (RSS, Reddit, industry newsletters) and scores them for engagement potential.
  • Strategist agent: picks the top topic and writes a carousel brief (hook, slides, CTA) using your brand voice guidelines.
  • Builder agent: generates the slide content and hands off to your design tool (Alex uses a Carousel Studio built on HTML/CSS + Puppeteer).
  • Delivery loop: runs on a cron schedule on the Mac Mini; finished assets dropped into a review folder for one-tap approval.

The key unlock: Claude Code's sub-agent feature. One orchestrator agent spawns cheaper specialist agents for each step.

Claude Code docs →
∞
LOOP
Build

Loops, Not Prompts

Build loops that run until done — unattended.

Not a tool — a mindset shift that unlocks the entire arsenal. A loop runs a task until it's actually done, unattended. You stop babysitting one-shot prompts and start setting goals that agents run to completion.

THE PATTERN

  1. Define "done" in concrete terms before you start (e.g., "generate 10 carousel posts, each with 8 slides, saved to /output/").
  2. Write a loop prompt in Claude Code: "Keep working until all 10 are complete. Don't stop to check in."
  3. Set guardrails: tell the agent what to skip, what to flag for human review, and where to save output.
  4. Walk away. Come back to the result.

In Claude Code this is the --dangerously-skip-permissions flag or simply writing your prompt to be self-directing.

Claude Code docs →
/CMD
Build

23 Skills · 19 Commands

The whole agency in one keystroke.

Alex's library of custom Claude Code commands — /discovery, /proposal, /operator, /books, and more. Each command is a repeatable workflow that executes an entire SOP in one keystroke. The agency's processes, made executable.

THIS IS A PROPRIETARY SYSTEM

Not public installs. Here's how to build your own:

  • In Claude Code, custom commands live in ~/.claude/commands/ as markdown files. Each file is a command.
  • Write your SOP as a detailed Claude prompt inside the markdown file.
  • Save as proposal.md → run with /proposal in Claude Code.
  • Start with your highest-friction recurring task. Write the full SOP, paste it as a command, test until it produces the right output every time.

Official docs on custom commands: Claude Code slash commands →

Custom commands docs →
Build

n8n

Visual automation without the SaaS lock-in.

An open-source workflow automation tool. Alex self-hosts n8n on his Mac Mini for local-first, private automations that connect tools, trigger agents, and run scheduled workflows.

WHAT YOU NEED

Node.js installed, or Docker. Self-hostable, or use n8n Cloud.

SETUP

  1. Self-host with npm: npx n8n — runs instantly on localhost.
  2. Or use Docker: see n8n Docker install docs.
  3. Open http://localhost:5678 to access the editor.
  4. Create your first workflow by connecting trigger → action nodes.
  5. Connect to Claude via the built-in AI nodes or HTTP request nodes.
n8n.io →
Build

Supabase

The database and backend that powers Skyline OS.

An open-source Firebase alternative. Alex uses Supabase as the database backend for his custom CRM (Skyline OS) — storing leads, clients, deals, and agent outputs.

WHAT YOU NEED

A Supabase account. No server setup required.

SETUP

  1. Sign up at supabase.com .
  2. Create a new project and get your Project URL + anon key from Settings → API.
  3. Install the JS client: npm install @supabase/supabase-js
  4. Initialize: createClient(SUPABASE_URL, SUPABASE_ANON_KEY)
  5. Use the Table Editor to create your schema (leads, clients, deals tables).
  6. Enable Row Level Security and set policies for your access patterns.
Supabase docs →
Build

Vercel

Deploy in seconds. Auto-deploy on every push.

The hosting platform Alex deploys every site to. GitHub push → live in seconds. Handles CDN, SSL, preview URLs, and serverless functions automatically.

WHAT YOU NEED

A Vercel account and a GitHub repo with your project.

SETUP

  1. Sign up at vercel.com and connect your GitHub account.
  2. Click "Add New Project" → import your repo.
  3. Vercel auto-detects Next.js, React, and static sites — configure if needed.
  4. Click Deploy. Your site is live at yourproject.vercel.app.
  5. Every subsequent push to your main branch triggers an auto-deploy.
  6. Add a custom domain in Project Settings → Domains.
Vercel docs →
Ops

Ops

The systems that run the agency without constant human intervention.

Ops

Skyline Command Center

The cockpit that runs the whole agency.

Alex's custom-built agency OS — a single dashboard for every lead, deal, onboarding, support email, and follow-up. Built on GoHighLevel as the CRM hub and extended with Supabase for custom dashboards. Replaces 4–5 standalone SaaS tools.

THIS IS A PROPRIETARY BUILD — HERE'S THE PATTERN

  • Start with GoHighLevel as the CRM spine: contacts, pipelines, and automations. gohighlevel.com
  • Add a custom dashboard using Supabase + Next.js to display real-time KPIs (MRR, pipeline, leads) in one view.
  • Wire in your agents: Claude Code agents read from and write to Supabase, keeping the dashboard updated as work happens.
  • Document every status: every contact must have a status (lead / prospect / active / churned) and a next action. No orphan records.
GoHighLevel →
Ops

Sub-Agent Orchestration

One brief → a fleet of agents working in parallel.

Alex's approach to delegating work inside Claude Code: one orchestrator agent spawns cheaper, specialized sub-agents for research, building, and review. Parallel execution means tasks that would take hours happen in minutes.

THE PATTERN

  • Write an orchestrator prompt that defines the job, then explicitly tells Claude to spawn sub-agents for each phase.
  • In Claude Code, sub-agents run via the Task tool — the orchestrator creates tasks and delegates.
  • Assign the right model to each task: use cheaper/faster models for research and boilerplate; reserve Opus for reasoning and decisions.
  • The orchestrator collects results and assembles the final output.

Official sub-agent docs: Claude Code sub-agents →

Sub-agents docs →
OPEN
CLAW
Ops

OpenClaw

A 24/7 agent with a heartbeat, running local.

Alex's always-on agent runtime, installed on a Mac Mini. OpenClaw runs automations through nights and weekends, local-first — no cloud subscription required, data stays on owned hardware. He wakes up to work that already ran.

THIS IS A PROPRIETARY SYSTEM

OpenClaw is not publicly available. The pattern to replicate it:

  • Get an always-on machine (Mac Mini, a VPS, or a Raspberry Pi).
  • Install Claude Code CLI on it and authenticate.
  • Set up cron jobs (or launchd on Mac) to fire Claude Code tasks on a schedule.
  • Use a local-first approach: keep sensitive data on the machine, not in a cloud AI service.
  • Connect via Tailscale for secure remote access without port-forwarding.
Tailscale (for remote access) →
Gmail Google Calendar Google Drive Notion Vercel
Ops

11 Connected MCPs

Claude that acts — not just chats.

Alex has 11 MCP servers wired into Claude Code. The most-used: Gmail, Google Calendar, Google Drive, Notion, and Vercel. Claude books meetings, sends emails, creates files, and ships deploys — all from one chat interface.

WHAT YOU NEED

Claude Code CLI installed (see Start Here). OAuth credentials for each service you want to connect.

SETUP

  1. Install Claude Code CLI: npm install -g @anthropic-ai/claude-code
  2. Browse available MCP servers at modelcontextprotocol.io or the Claude Code MCP directory.
  3. For Google services (Gmail, Calendar, Drive): create a Google Cloud project, enable the relevant APIs, and generate OAuth 2.0 credentials.
  4. Run claude mcp add and follow prompts, or manually add server config to ~/.claude/config.json.
  5. Restart Claude Code and verify: ask Claude to "list my recent emails" — if it returns results, Gmail MCP is live.
MCP registry →
WISPR
FLOW
Ops

Wispr Flow

Run your whole stack by voice.

System-wide AI dictation for Mac. Works in every app — not just one. Automatically cleans and formats speech as you talk. Alex writes prompts, emails, and docs completely hands-free. ~3–4× faster than typing.

WHAT YOU NEED

A Mac (macOS), a Wispr Flow account, and a microphone.

SETUP

  1. Download from wisprflow.ai.
  2. Install and grant microphone + accessibility permissions (required for system-wide dictation).
  3. Set your activation shortcut (e.g., hold a key to dictate).
  4. Click into any text field in any app and start talking.
  5. Wispr automatically cleans filler words, adds punctuation, and formats output.
wisprflow.ai →
Intel

Intel

The knowledge systems that keep the whole operation sharp.

Intel

Claude Memory

A second brain that never forgets.

Claude's persistent memory of Alex's context, projects, and preferences. Every session starts already caught up — no re-briefing, no "let me explain the business again." It compounds: every conversation makes the next one faster.

WHAT YOU NEED

A Claude subscription (any plan). Memory is built into the Claude Code CLI via CLAUDE.md files.

SETUP

  1. In Claude Code, memory lives in ~/.claude/CLAUDE.md (global) and ./CLAUDE.md (per-project).
  2. Write your context into ~/.claude/CLAUDE.md: who you are, your business, your preferences, your active projects.
  3. Claude reads this file at the start of every session — it already knows everything.
  4. Update it as your context changes. Treat it as a living document.
  5. Claude.ai's web/mobile interface also has a separate memory feature you can enable in Settings.
claude.ai →
Intel

DeepSeek

Understand anything technical — for pennies.

Alex's on-demand explainer for deep technical topics. Breaks down code, research papers, and complex systems in plain English. A lightweight, fast model for pure explanation tasks.

WHAT YOU NEED

A DeepSeek account. API access for programmatic use; web interface for ad-hoc.

SETUP

  1. Sign up at platform.deepseek.com.
  2. New accounts receive free credits on signup.
  3. For API access: generate an API key in the dashboard.
  4. The API is compatible with the OpenAI SDK format — easy to integrate into existing code.
  5. For web use: chat.deepseek.com — paste the scary thing, get it explained.
platform.deepseek.com →
Intel

Vault Brain

The whole agency in one source of truth.

Alex's master knowledge base — every SOP, client, offer, and system mapped in one Obsidian vault, synced to iCloud. Every agent reads from the same source of truth. The business runs on documented systems, not memory.

THIS IS A PROPRIETARY SYSTEM — HERE'S THE PATTERN

  • Pick one tool for your vault: Obsidian (free, local-first) or Notion (cloud-native, good for sharing). Alex uses Obsidian + iCloud sync.
  • Create a folder structure that mirrors your business: clients, offers, SOPs, finance, marketing, tech-stack.
  • Write every SOP as a document in the vault — not in someone's head. If it's not written, it's not a system.
  • Point your agents at it: agents read vault files via the filesystem (Claude Code) or Notion MCP. They already know the business before the session starts.
Obsidian →
DAILY
READ
Intel

Daily AI Newsletter

Signal, not noise — stay first on every new tool.

Alex's filtered daily read that keeps him first on new tools and techniques. Ten minutes each morning. Being early on tools like Claude Code, MCPs, and Higgsfield compounds into a real comptetive edge.

THE APPROACH

  1. Pick 1–2 high-signal AI newsletters and read only those (avoid 10+).
  2. Curated options: The Rundown AI, TLDR AI, Ben's Bites, or follow specific researchers on X for raw signal.
  3. Block 10 minutes in the morning before opening anything else.
  4. When you see a new tool, open it immediately and spend 15 minutes testing. First-mover advantage is real in this space.
TLDR AI (free) →