Hey Prompt Warrior,

There's a quiet shift happening in how AI agents talk to other software. For a while it was APIs, then MCPs. Now it's CLIs, and even Peter Steinberger (the creator of OpenClaw) is pushing them hard.

The short version: a CLI is faster and more token efficient for an agent than an MCP doing the same job. In a side by side test I ran, the same task took 4 round trips with MCPs and 2 to 3 with the CLI.

In this post, I'll walk you through what CLIs actually are, why agents like them, and the 7 CLIs I use to 10x my Claude Code setup.

If you prefer watching over reading, here's the YouTube video.

Here's what we'll cover:

Read time: 6 minutes

If you're an agency owner, coach or consultant doing $10K-$100K/mo, I'd love to help you deploy AI agents in your business.

What you get:

Spots are limited. If that sounds like you, book a call here.

Now, back to the tutorial.

CLI stands for command line interface. It's a program you run by typing text commands in your terminal. No buttons, no windows, no mouse. Text in, text out.

CLIs have been around since the 1960s. Graphical interfaces got bolted on top later, for humans. The comeback is happening because agents don't need the graphical part. They just need the fastest way to take an action.

Here's the ladder:

Most software companies are building both right now, because it's still undecided which one wins. But for day to day work on your own machine, the CLI usually gets the same job done faster.

One real downside: CLIs are local only. They live on your machine. If you want a remote routine (one that runs in the cloud on a schedule), the CLI isn't installed there, so you'd have to install and authenticate it on every run, which turns it back into an API with extra steps. For remote jobs, MCPs are easier. For everything local, CLIs win.

Gives Claude Code access to your entire Google Workspace: Drive, Sheets, Gmail, Calendar, Docs, Slides, Tasks.

Best for: anything in Google. Sorting a sheet, writing and sending emails, drafting docs, organizing Drive folders.

Setup: Google "GWS CLI", grab the GitHub link, paste it into Claude Code and say install this. It's the most involved of the bunch because you do a manual OAuth setup in Google Cloud Console (you also have to enable each service you want, like Sheets and Gmail). Your agent walks you through it.

Real example: I dropped in a sheet and asked Claude to sort it alphabetically by service name. It ran gws sheets spreadsheets get, performed the sort, done. There's no Google Sheets MCP at all, so the CLI gives you more power here, not just more speed.

The best part: GWS can actually send emails. The Gmail MCP only creates drafts. GWS bypasses that.

Generates images and videos from Claude Code.

Best for: thumbnails, ad creative, UGC videos, anything visual.

Setup: the easy kind. Copy the URL, paste it, say install. Run the auth, connect in the browser tab that pops up, done. It also installs a few Higgsfield skills for you.

Real example: I asked it to make a YouTube thumbnail for this exact video using my profile pic from my asset folder. Took about 5 minutes (it picked the wrong skill once and I redirected it to the Higgsfield generate skill). The result was genuinely good. Same flow works for video, using one of my saved ad scripts and a Higgsfield avatar.

Pro tip: point it at the right skill explicitly. It sometimes grabs a neighboring skill on the first try.

Lets Claude control your browser.

Best for: anything that only exists behind a browser, with no CLI, MCP, or API.

Setup: the easiest one. Paste the URL, say install. No account, no auth. It just works.

Real example: I asked it to draft a post in my Skool community in a visible browser (not headless) and leave it for review. It opened the browser, navigated Skool, opened the post modal, typed the title and body. You can run it visible or headless depending on whether you want to watch.

(I did a whole video on browser automation if you want to go deeper on this one.)

Scrapes data from the internet using "actors", little modules built for specific sites.

Best for: lead lists, competitor research, any structured data pull.

Real example: I had Claude run a small test scrape of car dealerships in Orange County via the Google Maps actor. Got 10 results back in a table. You can chain actors together: scrape Maps, then hit each result's website or Facebook page to pull email addresses, and enrich a cold outreach list end to end.

Schedules posts across a bunch of social platforms (Instagram, TikTok, YouTube, and more).

Best for: killing the download-reupload-schedule loop.

Real example: I pasted a Google Drive video link and asked Postiz CLI to schedule it to Instagram an hour out. Because my other tools are connected, it downloaded the video from Drive first, then uploaded and scheduled it through Postiz. No manual export, no scheduling app, no per-platform busywork.

Reads and writes WhatsApp messages.

Best for: a busy WhatsApp you want help managing.

Real example: I asked it to scan my WhatsApp and tell me which chats need a response right now. It came back with a "respond today" list plus a table of everything else still open. It can also send messages (with a confirm step first), and download and transcribe voice messages.

Summarizes articles, web pages, YouTube videos, slides, screencasts.

Best for: fast context. I use it constantly to summarize YouTube videos.

Real example: I pointed it at my own browser automation video. It recognized it was mine, then gave back an accurate breakdown of the 3 methods covered. Chain it with other tools and you've got a solid summarization step in any workflow.

I didn't have time to cover these in full, but they're on my radar:

✅ CLIs are becoming the dominant way for agents to talk to software, ahead of MCPs and APIs

✅ They're leaner, more token efficient, and agents are good at using them (4 round trips vs 2-3 in my test)

✅ Most install in one step: paste the URL, say "install this"

✅ The one catch: CLIs are local only, so remote cloud routines still favor MCPs

✅ Printing Press can generate a CLI for almost any tool, so expect a lot more of these

Thanks for reading!

If you enjoyed this, consider forwarding this newsletter to a friend or colleague.

What would you like me to write about in the next post? Reply to this email and let me know.

Moritz

P.S. Installing and wiring up CLIs like these across a real business is exactly what I do inside AgentOS. If you want the whole system built for you instead of figuring it out tool by tool, take a look here.