The question i have is how to leverage MCP / Claude API without having to run the desktop app? For instance, what if we want to use the same tools but on a command line interface. instead? Perhaps we could use them several instances at the same time? How is that different from "tool use" on AI SDK by vercel for instance? It seems to have a lot of overlap..
MCP is an attempt to creat an open standard for tool use that provides well defined interfaces for client, server and LLM function calling. It is designed so that you can use custom solutions for any of those three elements. You can use an existing server in your own client such as your command line tool. You could replace the Claude backend with ChatGPT or a local model, etc. What most people are demonstrating right now is just how easy it is to create and modify servers. That alone is pretty impressive, but considering it’s only a third of what MCP offers MCP is a very exciting new bit of kit.
@@JoshuaKolden absolutely! i wonder if we could still use claude as backend and MCP tools but instead use a command line tool so we don't have to use the mouse and keep clicking "allow" for instance. any ideas on how to do that? i saw documentation on how client and server works but not how to have an AI such as claude provide the clients and let it use the clients from a script as opposed to the desktop GUI.
Could you try forking one of the server repos? Then have Claude read the file and add additional tools to allow it to do jobs it doesn't yet have the tools for?
Great video & fab channel. However, the LLMs are only really good for small amounts of code. Once your project becomes complex or large, you get duff code generated all the time. So i think you will spend more time in cursor than in Claude desktop. If you have cursor setup with a github account, create a brand new project and then just push to github via the menu. It will just ask you for public or private. Then in composer ask it to create your application. Finally in composer ask it to create your readme file. Push the results. Although using MCP is interesting i dont see you gaining anything. A different example with something that saves time or is not doable in cursor would be good.
I made so called Server Commander before for ChatGPT CustomGPTs It gave terminal access to ChatGPT as action to call. That allows to run commands including deleting files or running things like tests.
@@freddiechipreswell if it can read and write to the software - it would be able to view, review, debug, edit and write automation flows for you. N8N flows are basically Json under the hood - så it wouldn't be that hard. If you export a flow and upload it to Claude it understands it pretty deeply already. This would just be that OK steroids.
I'm on the free plan and the only one I've been able to get to work is the SQLite server. I've tried PostgreSQL and GitHub with no luck. I'm getting this in the log file so not sure what is going on - Connected to MCP server postgres! and Connected to MCP server github!
Why dafuq the entheopic desktop app for linux does not exist while 80% of the software developer community and almost all 100% of the servers run on Linux
You're doing a fantastic job! Just a quick off-topic question: I have a SafePal wallet with USDT, and I have the seed phrase. (alarm fetch churn bridge exercise tape speak race clerk couch crater letter). How should I go about transferring them to Binance?
Adding a complex and error-prone middleware layer on top of GitHub's already user-friendly interface and basic Git commands isn't making things easier - it's creating unnecessary complexity and dependency. Instead of helping new developers learn essential Git skills they'll need in real-world team environments, this setup encourages avoiding those fundamentals entirely. The tutorial even casually dismisses implementation errors as normal behavior. As someone who values both AI and proper development practices, I'd strongly recommend new developers invest time in learning Git basics through GitHub's excellent documentation and UI instead of relying on unnecessary complexity. Good AI implementation should enhance our capabilities, not substitute for fundamental skills.
It can't possibly be that this is an early alpha technology and the very concept of developer will be entirely different.. no no technology will not progress. Everyone will stay the same.. hundreds of years understanding GitHub
Bei mir wurde nur der Titel auf deutsch angezeigt was ich auch schon seltsam finde… Man kann es aber sicher irgendwie deaktivieren. RUclips scheint da grade sehr viel rum zu experementieren
Oh I get this MCP thing now lol. Thanks for explaining it so well.
The question i have is how to leverage MCP / Claude API without having to run the desktop app?
For instance, what if we want to use the same tools but on a command line interface. instead?
Perhaps we could use them several instances at the same time?
How is that different from "tool use" on AI SDK by vercel for instance? It seems to have a lot of overlap..
MCP is an attempt to creat an open standard for tool use that provides well defined interfaces for client, server and LLM function calling. It is designed so that you can use custom solutions for any of those three elements. You can use an existing server in your own client such as your command line tool. You could replace the Claude backend with ChatGPT or a local model, etc.
What most people are demonstrating right now is just how easy it is to create and modify servers. That alone is pretty impressive, but considering it’s only a third of what MCP offers MCP is a very exciting new bit of kit.
@@JoshuaKolden absolutely! i wonder if we could still use claude as backend and MCP tools but instead use a command line tool so we don't have to use the mouse and keep clicking "allow" for instance.
any ideas on how to do that?
i saw documentation on how client and server works but not how to have an AI such as claude provide the clients and let it use the clients from a script as opposed to the desktop GUI.
How much do the API calls total up costing by the end of these projects?
That's the good part, because it's part of the Claude subscription already so you don't pay anything else for api calls to Claude 🙂
If you running Claude hard you could probably do 10-20/day. That’s doing a crazy amount of shit though. Really it’s like 1-5
when i did the git issue problem it failed every time saying its unable to access the repo despite it already created the repo and has all access
Your plans are top notch❤.
Hum Hindi mein ise sun pare yahi bahut behatareen hai!
Could you try forking one of the server repos? Then have Claude read the file and add additional tools to allow it to do jobs it doesn't yet have the tools for?
I would love to test out your version.
If you would add it somewhere please let me know.
claude desktop app runs into MCP fatal error (github api) and limit restrictions.
Would be great to make an agent that can play Pokemon TCG
Great video & fab channel.
However, the LLMs are only really good for small amounts of code. Once your project becomes complex or large, you get duff code generated all the time. So i think you will spend more time in cursor than in Claude desktop.
If you have cursor setup with a github account, create a brand new project and then just push to github via the menu. It will just ask you for public or private. Then in composer ask it to create your application. Finally in composer ask it to create your readme file. Push the results.
Although using MCP is interesting i dont see you gaining anything.
A different example with something that saves time or is not doable in cursor would be good.
I've not had any luck creating a MCP other than the SQL example running on Windows
do you run it on MacOS?
We NEED custom MCPs. I'll make one soon, but basically...you can't delete files (or git rm)
I made so called Server Commander before for ChatGPT CustomGPTs
It gave terminal access to ChatGPT as action to call.
That allows to run commands including deleting files or running things like tests.
How do I get MCP on my Claude account ?
Watch the video above this one
Master Control Program
I cannot wait for an N8N MCP
What would that look like?
wont happened
@@freddiechipreswell if it can read and write to the software - it would be able to view, review, debug, edit and write automation flows for you.
N8N flows are basically Json under the hood - så it wouldn't be that hard. If you export a flow and upload it to Claude it understands it pretty deeply already.
This would just be that OK steroids.
@@pushingpandas6479i am pretty certain it will? Why wouldn't it?
I wonder how much users on free plan can do with this
I'm on the free plan and the only one I've been able to get to work is the SQLite server. I've tried PostgreSQL and GitHub with no luck. I'm getting this in the log file so not sure what is going on - Connected to MCP server postgres! and Connected to MCP server github!
Why dafuq the entheopic desktop app for linux does not exist while 80% of the software developer community and almost all 100% of the servers run on Linux
Where did you pull that statistic from?
You're doing a fantastic job! Just a quick off-topic question: I have a SafePal wallet with USDT, and I have the seed phrase. (alarm fetch churn bridge exercise tape speak race clerk couch crater letter). How should I go about transferring them to Binance?
Why is windows such a hassle
A lot of flaws so far...
Aru u making money of ai?
@@firemontana1284 are you? I want to. I'm trying.
❤
There's a lot of money to be made from an app that does this reliably...
Adding a complex and error-prone middleware layer on top of GitHub's already user-friendly interface and basic Git commands isn't making things easier - it's creating unnecessary complexity and dependency. Instead of helping new developers learn essential Git skills they'll need in real-world team environments, this setup encourages avoiding those fundamentals entirely. The tutorial even casually dismisses implementation errors as normal behavior. As someone who values both AI and proper development practices, I'd strongly recommend new developers invest time in learning Git basics through GitHub's excellent documentation and UI instead of relying on unnecessary complexity. Good AI implementation should enhance our capabilities, not substitute for fundamental skills.
It can't possibly be that this is an early alpha technology and the very concept of developer will be entirely different.. no no technology will not progress. Everyone will stay the same.. hundreds of years understanding GitHub
i got this video translated in german? terrible, why?
Bei mir wurde nur der Titel auf deutsch angezeigt was ich auch schon seltsam finde…
Man kann es aber sicher irgendwie deaktivieren. RUclips scheint da grade sehr viel rum zu experementieren
@ zweiteres dachte ich mir auch schon, hab aber nichts gefunden …
@ danke für dein Input, sind wir jetzt Freunde? 😃
@@youtubeccia9276 aber natürlich 😁
@@fabiankliebhan feine sache :)
Wispr Flow… is just the best