20:04Chase AI · Tutorial
Claude Code Has Evolved
Why the skill backbone — not the dashboard — is where all the real value in a Claude Code Agentic OS lives.
May 14thThe argument in one line.
The terminal is Claude Code native environment, and CLI tools that live there beat MCP servers on token efficiency and setup speed, sometimes by 90000 tokens per task.
Read if. Skip if.
READ IF YOU ARE…
- You use Claude Code daily and keep switching to browser tabs to manage Stripe, Vercel, or GitHub.
- You have at least one MCP installed and wonder whether it is actually saving you tokens or costing you more.
- You do any video research or media processing and want Claude Code to handle more of that pipeline.
- You are running local LLMs and struggle to know which Ollama model actually fits your hardware.
SKIP IF…
- You are not using Claude Code or another terminal-based coding agent yet.
- You are looking for deep setup guides. This video names tools and links repos but does not walk through installation step by step.
The full version, fast.
CLIs are replacing MCPs as the default extension layer for Claude Code because they run in the same terminal, avoid schema loading overhead, and use dramatically fewer tokens. This video covers 10 tools: CLI Anything, NotebookLM-py, Stripe CLI, FFmpeg, GitHub CLI, Vercel CLI, Supabase CLI, Playwright CLI, LLMFit, and gws. The core advice: install the CLI as a dependency, then add its companion skill file so Claude Code knows how to drive it.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:27
01 · Intro
Host frames the shift from MCPs to CLIs as the dominant trend. Promises 10 tools spanning research to deployment.
00:27 – 02:18
02 · CLI Anything
Open-source tool that auto-generates production CLIs for any open-source software. One command runs the full pipeline.
02:18 – 04:38
03 · NotebookLM-py
Unofficial Python API for NotebookLM. Lets Claude Code throw YouTube URLs at NotebookLM for free video analysis then pull back deliverables programmatically.
04:38 – 05:23
04 · Stripe CLI
Manages Stripe products webhooks and events from the terminal. Eliminates 20-tab browser navigation for payment product setup.
05:23 – 06:30
05 · FFmpeg
Audio/video/subtitle manipulation library. Demonstrated by chopping a keyboard product video into frames for a hero scroll animation.
06:30 – 07:30
06 · GitHub CLI
Native GitHub operations from the terminal. Claude Code already understands it well enough to install and authenticate in one sentence.
07:30 – 08:38
07 · Vercel CLI
Deployment and CI/CD pipeline from the terminal. Vercel publishes an official agent skills library with categories including deployment browser automation and design.
08:38 – 09:02
08 · Supabase CLI
Runs the full Supabase stack locally. Handles migrations auth and database management without a cloud connection.
09:02 – 11:00
09 · Playwright CLI
Browser automation CLI. Own benchmark showed CLI was 90000 tokens cheaper than the MCP equivalent. Skill installs in one line.
11:00 – 11:40
10 · LLMFit
TUI tool that scores every available Ollama model against your actual hardware specs and tells you which ones will run.
11:40 – 13:24
11 · gws Google Workspace CLI
Controls Gmail Docs Drive Sheets and Calendar from the terminal with 40+ agent skills. Reads Discovery Service at runtime so new APIs appear automatically.
13:24 – 14:03
12 · Final Thoughts
Host frames the CLI trend as structural: CLIs and Claude Code share the terminal natively so there is no overhead.
Lines worth screenshotting.
- Playwright CLI completed the same browser task as its MCP counterpart using 90000 fewer tokens.
- NotebookLM-py offloads video analysis to Google servers so those tokens bill against Google not your API key.
- CLI Anything can generate a full production CLI for any open-source software in a single command with no API required.
- gws reads Google Discovery Service at runtime so every new Workspace API endpoint becomes available automatically without updating the CLI.
- Almost every CLI in this list follows the same two-step pattern: install the dependency then add the companion skill file.
- LLMFit scores every available Ollama model against your actual RAM GPU and CPU so the right model is a TUI selection not a guess.
- Vercel publishes an official skills library with pre-built agent behaviors for deployment browser automation commerce and design.
- The shift from MCPs to CLIs is structural: Claude Code lives in the terminal and zero-overhead direct connections beat schema-loaded plugin servers for throughput.
- FFmpeg gives Claude Code the ability to chop video into frames loop and stitch animations and extract audio with no native path otherwise.
- The Supabase CLI lets you run the entire Supabase stack locally so database migrations and auth can be developed without a cloud connection.
- Stripe CLI replaces 20-tab browser navigation with terminal commands for product creation webhook testing and event tailing.
- The GitHub CLI is so well understood by Claude Code that you can install and authenticate it in a single natural-language instruction.
CLIs beat MCPs where it costs most: tokens.
WHAT TO LEARN
The terminal is already where Claude Code lives, and tools that meet it there skip the entire schema-loading overhead that makes MCP servers expensive.
01Intro
- The shift from MCPs to CLIs is structural not a preference: CLIs share the terminal with Claude Code and have zero protocol overhead.
02CLI Anything
- CLI Anything auto-generates a full production CLI for any open-source software in one command with no API no wrappers just structured JSON output Claude Code can drive.
03NotebookLM-py
- NotebookLM-py lets you offload video and document analysis to Google servers for free then pull the results back into your Claude Code session programmatically.
- The two-step pattern applies here: install the Python package then add the companion skill file so Claude Code knows the commands and output formats.
04Stripe CLI
- The Stripe CLI collapses multi-tab browser workflows for product creation webhook testing and event tailing into single terminal commands.
05FFmpeg
- FFmpeg gives Claude Code direct access to video manipulation including chopping looping reversing and stitching, capabilities that have no native path in any current AI coding agent.
06GitHub CLI
- Claude Code understands GitHub operations well enough that installing and authenticating the GitHub CLI requires one sentence and it handles the rest including the OAuth browser click.
07Vercel CLI
- Vercel publishes an official agent skills library. Check it before writing custom skill files since pre-built behaviors exist for deployment browser automation commerce and design.
08Supabase CLI
- The Supabase CLI lets you run the full stack locally so database migrations and auth flows can be developed and tested without a live cloud project.
09Playwright CLI
- Playwright own benchmark showed its CLI completed the same browser task as the MCP server using 90000 fewer tokens, the clearest published data point on the CLI vs MCP tradeoff.
- Playwright CLI goes deep: beyond form testing it supports session storage network mocking DevTools tracing and video recording of browser sessions.
10LLMFit
- LLMFit removes the guesswork from local model selection by scoring every Ollama model against your actual hardware specs and flagging which ones are marginal good or infeasible.
11gws Google Workspace CLI
- gws gives Claude Code access to the entire Google Workspace API surface with 40+ pre-built agent skills and security sandboxing via scope filters.
- Because gws reads Google Discovery Service at runtime new Workspace API endpoints appear automatically without updating the CLI itself.
12Final Thoughts
- The throughput and token efficiency advantage of CLIs over MCPs compounds as you add more tools, each saved schema load and avoided context prefill adds up across a full workday.
Terms worth knowing.
- CLI (Command-Line Interface)
- A text-based program invoked from a terminal. Claude Code calls CLI tools directly as shell commands with no additional protocol layer.
- MCP (Model Context Protocol)
- Anthropic plugin protocol for adding tools to Claude. MCPs load their full schema upfront which costs tokens before any task begins.
- Skill file
- A markdown file placed in Claude Code context that teaches it how to use a specific CLI tool including the conventions flags to prefer and outputs to expect.
- Discovery Service
- A Google API endpoint that lists all available Workspace APIs at runtime. gws uses it to build its command surface dynamically.
- LLMFit
- An open-source CLI tool that benchmarks your machine RAM CPU and GPU against available Ollama models and recommends which ones will actually run.
Things they pointed at.
Lines you could clip.
09:17
“CLI, essentially, to do the exact same thing as the MCP server was both quicker and it was like 90000 less tokens.”
13:38
“We are moving away from MCPs. We are moving into CLIs because it just makes sense. Claude Code lives in the terminal. CLIs live in the terminal. There is no overhead.”
02:42
“I can just throw YouTube URLs at NotebookLM. It will do all the analysis for me for free because these tokens are on Google servers not ours.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
00:00If you've been paying attention to the Cloud Code ecosystem at all, then you've noticed a huge shift. Everyone, I mean everyone, is building CLI tools to enhance what Cloud Code can do. And today, I'm gonna show you my 10 favorites, and they range from YouTube research to deploying applications to controlling my entire Google Suite, all from Cloud Code.
00:21And you might recognize a few of the ones we cover today, but if you leave with just a couple new ones, then I'll have done my job. So the first tool on our list is kind of a meta pick, and that is CLI Anything.
00:31This is a CLI tool that creates other CLI tools. This thing is completely open source and it's from the makers of LightRag and Rag Anything. So these guys are kind of titans in the AI open source world.
00:43Now the sell for this particular tool is that I can use CLI anything and point Cloud Code at any open source project. And Cloud Code will use CLI anything to create a CLI tool for said program. And this repo shows some of the projects that they've already done this on.
01:00Things like GIMP, Blender, InkSpace, OBS, Zoom.
01:05We actually did a full deep dive on this exact tool not too long ago. So I'll put a link to that above. And it looks like they've added a few since that NotebookLM as well.
01:13Right? So the idea is if you want Claude code to work with some other program from the terminal and there's not like a clear API type deal to go along with it and it's open source, well, this is the perfect tool for that job. And it's also really easy to install and use.
01:28It's like two steps to install it, and then it's just a one step process to actually execute this. It runs the full pipeline on its own. And for you, the user, it's very, very simple.
01:38So before we jump into tool number two, a quick word from everybody's favorite sponsor, me. I just released the Claude code masterclass.
01:46You can find it inside of chase.aiplus. There is a link to that in the pinned comment. And it is the perfect place if you're trying to figure out how can I go from zero to AI dev, especially if you don't come from a technical background?
01:57I add updates to this every single week, and we really focus on real life practical use cases, not just feature explanations. So if that's something you're interested in, definitely check it out. And if you're brand new, I also have the free Chase AI community.
02:12Tons of free resources. Link to that is in the description, so hopefully there's something there for everybody. Now tool number two is one of my personal favorites and you've seen me showcase this before and that is the notebook l m dash pi tool.
02:25This tool, this CLI tool allows us to connect Cloud Code to NotebookLM. There's no public API for NotebookLM. So there's no way normally for me to control NotebookLM from the terminal unless I've created some sort of custom browser automation.
02:39But this GitHub repo fixes that, and this is something I quite literally use every day for my own research because it solves one of the issues with Claude code and Sonnet and Opus in general. It's the fact that they can't really handle videos.
02:52NotebookLM can. I can just throw YouTube URLs at NotebookLM. It will do all the analysis for me for free because these tokens are on Google servers, not ours, and then it just brings it back to Cloud Code.
03:03And on top of the analysis it does in Notebook LM, it can create all the deliverables that Notebook LM does as well. So things like podcasts, videos, slide decks, infographics, quizzes, flashcards, on and on and on.
03:15If you can do it in Notebook LM, you can do it inside of Cloud Code. And being able to control all of this from a terminal is a huge benefit versus doing it manually inside of NotebookLM.
03:24And it's not just a convenience thing. It's the fact that I can very easily take all the analysis from NotebookLM and integrate it into whatever I'm doing at a larger scale inside of my projects.
03:35And you also get access to things you can't really do inside of Notebook LM's web interface. And they're all listed down here, everything from batch downloads to slide revision to programmatic sharing.
03:45Now like many of these CLI tools, it's really a two step process because we need to, one, actually install the CLI tool, the dependency on our actual computer. And then number two, you know, we need a skill.
03:57Skills and CLI tools tend to go hand in hand when it comes to Claude code because we need to essentially give it the functionality and then we need to teach it how to use that specific tool in the specific way. So if we're talking about CLI tools that don't require a skill, I'll let you know.
04:13But this is one of them that does. And in pretty much all these repos, there will be some sort of command that you just have to copy paste to get the skills loaded. And also, you know, while all the install is spelled out for us here, there is nothing stopping you from just giving Code the URL to whatever repo we're talking about or, you know, whatever web page happens to explain the CLI, give it the Cloud Code and it will install it for you.
04:38Now I really like tool number three because it's all about money, and that's the Stripe CLI. Now if you've ever used Stripe to essentially create a product that's tied into something like a web app, you know what a pain in the butt it is to go through Stripe's interface.
04:52It is not user friendly at all. However, when we use the Stripe CLI, it takes all of that turmoil away from you and puts it on top of Cloud Code, and Cloud Code is able to navigate all that nonsense much more quickly.
05:05That being said, when you are dealing with things that have to do with money and transactions, like, obviously, you still wanna test these out by hand. But the actual, like, product creation and all, like, the steps that require you to go through, like, 20 different tabs inside of Stripe, like, we can just get rid of that.
05:20This makes it way easier. Now tool number four is also in this content slash multimedia video game, and that's FFmpeg. And this is a collection of libraries that allow you to manipulate video and audio and subtitles, and it actually can be very useful in things like web design.
05:36So I was able to create this scrolling animation on my web page with FFmpeg. And what it was able to do is it was able to take an entire video that I had created of this keyboard going from, you know, completely together to, like, completely expanded and then chop it up into individual frames to turn it into what you see here.
05:56And if I had tried to do that manually, it'd be giant pain in the butt. But something like FFmpeg makes this very simple. If I wanted to do a, let's say, an animation on the hero section, let's say I had some sort of image back here and I wanted to loop it, well, I could just give it one video.
06:09It can automatically copy the video, reverse it, stitch it. So it gives Claude code a lot of functionality when it comes to manipulating sort of multimedia, which is, again, it's a place that Claude code out of the box isn't great at, so these CLI tools that can give it some enhanced functionality in those spaces are always welcome.
06:27And I also have a full video breaking down how to do this exactly that I'll link above. Now tool number five is one you probably already know about and you definitely should be using. If not, you're way behind.
06:35And that is the GitHub CLI. If we are doing anything where we are writing code and we wanna push to GitHub, there is no reason why we wouldn't just use the GitHub CLI to do this. Right?
06:46Anytime where we have some sort of task or workflow that we are constantly doing and pushing to GitHub and anything on the deployment side is probably one of those, we wanna be looking for CLIs to help us do it. And the GitHub CLI, you really can't ask for anything better.
07:00Now like all things Git and GitHub related, Cloud Code is already very spun off on how to do this properly. So when it comes to installing the GitHub CLI and using it, quite literally, you can do that in one sentence. Open up Cloud Code and say, hey.
07:11I wanna install the GitHub CLI. It will do it automatically. It's just gonna ask you to authenticate, which is a really simple process where you click a link, you log in to GitHub, and you're all set.
07:20And from here, when it comes to pushing code, when it comes to, you know, commits and doing branches and all these things, you can just do it from the terminal. Now while we're on the topic of deployments, we would be remiss if we didn't talk about tool number six, Vercel CLI.
07:35Now I love Vercel because they have a very generous free tier, and their connection with GitHub makes it very simple to essentially create a CICD pipeline just with these two tools. Now installing the Vercel CLI is again very simple, and this just allows you, like all these CLI tools, to control some part of the architecture process, the deployment process from your terminal so not switching from tabs back and forth.
07:59Now if you're ever wondering, hey. How do I actually use this? Well, you know what I'm gonna tell you.
08:04Take this page, copy the whole dang thing, throw it inside of Claude code, and it won't know what to do. Now when it comes to skills with the Vercel CLI, Vercel actually has a ton of skills available for you to give to Claude code. But if we head over here to deployment, they have a specific one for Vercel CLI, and they have a few others that are in the deployment realm, specifically Vercel Deploy and Autoship.
08:24But I definitely suggest checking out this page. I'll put a link to that down below as well because they actually have a ton of really useful skills that if you're doing anything like browser automation as well, um, same thing like design and UI, you definitely wanna take a look at.
08:37Now tool number seven is all about the back end, and that's the Supabase CLI. I like Supabase for the same reason I like Vercel. Very generous, free tier, and I can handle both all of my databases and authentication from one place from the terminal using this tool.
08:52And remember, Supabase is also open source. It was always meant to be a Firebase alternative. So if you wanna run Supabase completely locally, the CLI tool helps you do that as well.
09:01Now tool number eight is all about browser automations and giving Cloud Code the ability to spin up its own instances of Chrome and do things on the web for you. And that's where Playwright CLI comes in.
09:12Playwrights Playwrights CLI is also a great case study when we look at CLIs versus MCPs. I highly suggest you take a look at the Playwright YouTube account.
09:21They have an entire video on this where they compared them head to head. And CLI, essentially, to do the exact same thing as the MCP server was both quicker and it was, like, 90,000 less tokens.
09:32And that's kind of what you kind of see here with all these CLI tools versus their MCP counterparts even with the changes that ClodgeCode has made when it comes to MCPs where everything isn't just front loaded. So definitely an interesting thing.
09:45But the Playwright CLI is great. If you want an alternative while we're on the, you know, topic of browser automations, Vercel actually has its own.
09:55Right? So I mentioned that a little bit before. The agent browser skill.
09:58So if you don't love Playwright CLI, you can always check out the Vercel alternative. But I highly suggest you start integrating something like Playwright CLI into your stack, especially if you're someone who builds web apps.
10:11Why is that? Because inevitably, we need to test things on the web apps we build, and I don't wanna go into a new tab every single time, you know, go on the dev server, check out the new form myself. Why don't I just have Cloud Code do that on its own, spin up five Chrome tabs, and just attack it from a bunch of different angles automatically?
10:28Right? This allows us to do that. Like almost all these, there is also the requisite skills we want to install here, and it's just a one liner, and it will automatically go inside your DocClot folder.
10:39Now I will say Playwright in particular actually is pretty deep. There's a whole lot of things you can do with it beyond just, hey, open up Chrome and test out this form submission. So if you're really into browser automation, I highly suggest you kind of dive into this repo and see what's really available to you because it again, these are deep waters when you go into the browser automation stuff.
10:57Now tool number nine is a pretty interesting one. I just got introduced to it not too long ago. You might have seen me reference it in my how to run Claude code locally video, and that's LLM Fit.
11:08This is a CLI tool that allows you when you run it to figure out, well, what local model actually makes sense for my setup? And for most people, that question has a non obvious answer. I mean, if you head to OLAMA and go to the models, like, it's essentially like a never ending list of different things to do and open source is being updated every three seconds.
11:27And so, you know, every single model also has, like, nine different versions of the same model. So it's it's kind of a confusing space, especially if it's not one you hang out in very often. And this project, LLM Fit, essentially solves that issue.
11:39And last but not least is GWS, the Google Workspace CLI tool. This is the CLI tool that allows us to have Claude code control our entire Google Workspace.
11:50Right? We're giving it access to our email, our docs, sheets, everything.
11:56Right? This is extremely powerful, but this also is the place where you start getting into those security issues. Like, do we necessarily want Code to have access to all our emails?
12:05But luckily, it's not too hard to set up the GWS CLI tool in a way where we almost, like, sandbox Cloud Code, and it has, like, shared folders with us, and it only has access to certain emails if we use filters. So there is some wiggle room if you're scared about the security side. But for many people, having Cloud Code have full access to the Google Suite is a big deal.
12:24And even if you do give Claude Code access to, say, your Google email, Google Workspace has something called Armor, and it's all about dealing with, like, prompt injections and that sort of thing.
12:34So there are guardrails in place. We aren't just, like, letting Claude code go off and do whatever it wants. Now this one in particular has, like, a billion skills.
12:43Like, I'm not really exaggerating. Right? So a big part of figuring out how to use this tool to the most effect is also figuring out which skills make the most sense.
12:53Right? Skills aren't a huge context window, you know, drag. But if you have too many of them, triggering the right one becomes a problem.
13:01So this is the kind of repo where what I would do and what I have done when I installed this thing was I essentially pointed Cloud Code at this thing. I created a new directory. I had it clone the whole thing.
13:10And then you can have a discussion with Cloud Code and say, hey, like, alright. Now that you have full visibility into how this works and what's going on, what would make the most sense for us to actually, like, install and use. Right?
13:21And that's gonna be different for every single person. So that's my list of my 10 favorite CLI tools as of today. Tomorrow, that list will probably change in a week.
13:28I'll probably have added, like, three more of them because I wasn't joking about it in the intro. Like, this is the shift. Like, we're moving away from MCPs.
13:36We're moving into CLIs because it just makes sense. Cloud Code lives in the terminal. CLIs live in the terminal.
13:41There's no overhead. It's, like, just a straight connection. It allows Cloud Code to do the most with the least amount of tokens for all intents and purposes.
13:50So hopefully at least a couple of these were new to you and you can begin to integrate them into your stack. So let me know in the comments what you thought. Make sure to check out Chase AI plus if you're interested in the master class, and I'll see you guys around.
The Hook
The bait, then the rug-pull.
Everyone is building CLI tools to extend Claude Code, and one benchmark from the Playwright team put the advantage in hard numbers: the same browser task that cost an MCP server a certain token budget cost the CLI equivalent 90000 fewer. That gap is why this list exists.
Named ideas worth stealing.
03:56model
Two-Step CLI Integration Pattern
- Install the CLI dependency on your machine
- Add the companion skill file so Claude Code knows how to drive it
Every CLI tool in the list follows this same installation pattern.
Steal forany onboarding doc or README for Claude Code integrations
09:13concept
CLI vs MCP Decision Rule
- CLI: prefer when throughput matters, task is frequent, tool lives in the terminal
- MCP: prefer when persistent browser context is needed or long-running autonomous loops require stateful reasoning
Playwright benchmark showed CLIs win on token efficiency. MCPs retain value for stateful long-running autonomous flows.
Steal forarchitecture decisions on any Claude Code stack
How they asked for the click.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
