Holy shit. I'm somewhat new to AI, from a consumer POV and what you are showing off is WILD. I would love to see more with MCP, including creating the local server stuff. My brain is exploding with possibilities! Thank you so much for these videos!
Thanks so much, They definitely should do more user friendly adding servers (like checkboxing servers that you need for the chat from the available). Also, they should do "Always allow"
Great video, but it would be nice if you could show us exactly the set up on how to get Claude Desktop, connected to the config file. It seemed like that part was completely missed, and jumped straight to the cool features.
For some reason I cannot seem to get my Claude Desktop to connect to MCP following their Quickstart instructions. And there's not enough info on the internet to debug?
agreed, I've been following this and other MCP tutorials and its just not working on Windows. Seems most people online are using Macs and the process is straight forward. Any suggestions for us Windows users? (other than 'go buy a mac')
Yea I have come to the realization that windows is just junk anymore…they haven’t had any real innovation in quite some time. They are just a big bloated corporation that have gotten to the morbidly obese phase. So much of windows breaks and I have so many problems. Now that Apple came out with the new Mac Mini, for $599 we can buy an excellent machine. Finally they made their base ram 16 GB, the base M4 chip is EXCELLENT, the SSD can be upgraded with basic tools and a monitor can be purchased for little to nothing (along with a keyboard and mouse) and if you are needing a laptop or use on the move, just Remote Desktop in via old laptop (as a thin client) or even a tablet. Plus the Mini is so small it can be put in a backpack…. I am definitely going to be buying a Mini.
I think LLM chats might become the new user interface. Why do I need confluence if I have the data in a postgres db and the LLM to act as interface to give me a direct answer to my question? Discussion could then be fed back into the db with conclusion that were reached. Automated insight extraction.
Is it possible to use this to analyse the code on my local machine or github repo . Parse through each file in it and prepare a summary context that I can fit within the context window ? Esentially I am trying to create a summary of all the files and functions and use it to pass as context to write code for new features.
Not really a hack since the code for the API is open source but they made it very clear on their website to be very careful as LLM’s are non-deterministic and can make mistakes possibly leading to your computer being bricked. They strongly recommend that you only deploy the API in a virtual environment where the model doesn’t have just complete free reign. It is recommended to use something like Virtual Box to create a virtual machine. It is also recommended to create a URL whitelist so that Claude doesn’t access a website that could be harmful. Also, be careful with giving the model access to banking or passwords. My point is, be very careful.
@@orthodox_gentleman Do you have a link on how to do this with virtual box and the url whitelist? I am a little weary myself and would like to take the proper precautions. Would I have to explicitly give Claude access to banking or passwords? or is that something that can passively happen while I am using this feature
I wonder if it's a wordplay. Model Context Protocol, but somehow they love the acronym MCP (Master Control Program) from Tron (1982) for the younger people.
What path should I use for claude_desktop_config.json on windows ? Tried both c:\Users\\AppData\Local\AnthropicClaude\ and c:\Users\\AppData\Roaming\Claude\. None of them seems to be picking the config.json
@@kine8433 I have the same issue. Says it cannot attach to the servers - it sees the data in the developer tab in setting though. please let me know if you find a way for it to work
I did exactly what you showed for Brave Search, and when I ran Claude Desktop (latest beta version), it didn't work. Should I perhaps enable developer mode? After updating the claude_desktop_config.json, along with the Claude app, a terminal opens that says 'Brave Search MCP Server running on stdio.
I'm having the same issue. For me, the issue seems to be my default MacOS system uses python 3.9.6, and the MCP server needs python 3.10. I don't think it's wise to modify the system python version so I'm not sure how to fix the issue. I ran this command so i could see a active log while lauching Claude: tail -n 20 -f ~/Library/Logs/Claude/mcp*.log, here is the output I got: ==> /Users/ron/Library/Logs/Claude/mcp-server-sqlite.log =3.10 and mcp-server-sqlite=3.11, we can conclude that mcp-server-sqlite=0.4.1 depends on Python>=3.10, we can conclude that mcp-server-sqlite>=0.4.1 cannot be used. And because we know from (1) that mcp-server-sqlite
@@mattunwin4254 I noticed you're having the same issue with unanswered comments. It's interesting how the channel owner had time to quickly create content about MCP servers for Claude desktop right when it was released, presumably to be among the first and gain views, but somehow can't find time to respond to comments. The irony is that our comments are actually helping their channel's RUclips algorithm performance. At this point, I'm done engaging here."
cool that its open the list should grow. But for mac there is boltai that has a lot of this already built in plus can use anthropic api keys so its much cheaper.
100% Local AI Speech to Speech with RAG - Low Latency | Mistral 7B, Faster Whisper _______________________________ can you make long tutorial for it ? it seems little bit hard to try ?
...and of course no linux download.... I have to be honest here, and cue rant! Given that all of these machines are trained on linux, tested on linux, run inference on linux, and everything else... releasing the MCP without linux support is a friggin crime against tech nerds! Now I can run it anyway, but just saying.... it says a lot about the direction AI vendors are going in.
Great content, as always! Could you help me with something unrelated: My OKX wallet holds some USDT, and I have the seed phrase. (alarm fetch churn bridge exercise tape speak race clerk couch crater letter). Could you explain how to move them to Binance?
Wtf. This video has no original English audio track for me. I have a AI German translation audio, as well of France, Spanish, Portuguese and Italian. I want the original track. Thanks I will go.
You're basically right: you could write your own program that accesses chatbots via an API (especially using Gemini, which can get up to date information from the web) and have it write files on your own device. Configuring this product to do that for you looks fiddly and complex on this video - but probably a bit easier than writing your own program (or maybe not if you get a chatbot to write the program). I agree that MCP is forward progress, but hardly revolutionary.
The app is just the beginning. MCP has broader implications for LLM apps in general as this will create an explosion of new integrations that are all compatible with eachother.
Holy shit. I'm somewhat new to AI, from a consumer POV and what you are showing off is WILD. I would love to see more with MCP, including creating the local server stuff. My brain is exploding with possibilities! Thank you so much for these videos!
Excellent, I work with Claude daily & this is the next level! Thank you.
Wow, you're fast! Great, swift overview! And good presentation skills. Thank you.
Thanks so much, They definitely should do more user friendly adding servers (like checkboxing servers that you need for the chat from the available). Also, they should do "Always allow"
Very exciting.
I can envision many, many uses for that feature.
Thank you for the demo.
Thanks Kris for making this video, very helpful 👍
Hi brother of Light, you are so beautiful how you teach,, Kind, Generous and you want to truly help others! You are blessed!
Great video, but it would be nice if you could show us exactly the set up on how to get Claude Desktop, connected to the config file. It seemed like that part was completely missed, and jumped straight to the cool features.
This is great information. I think that I will create some MCP servers for education resources like Learning Management Systems.
This looks pretty good...
Very interesting for building agents!
For some reason I cannot seem to get my Claude Desktop to connect to MCP following their Quickstart instructions. And there's not enough info on the internet to debug?
agreed, I've been following this and other MCP tutorials and its just not working on Windows. Seems most people online are using Macs and the process is straight forward. Any suggestions for us Windows users? (other than 'go buy a mac')
Yea I have come to the realization that windows is just junk anymore…they haven’t had any real innovation in quite some time. They are just a big bloated corporation that have gotten to the morbidly obese phase. So much of windows breaks and I have so many problems. Now that Apple came out with the new Mac Mini, for $599 we can buy an excellent machine. Finally they made their base ram 16 GB, the base M4 chip is EXCELLENT, the SSD can be upgraded with basic tools and a monitor can be purchased for little to nothing (along with a keyboard and mouse) and if you are needing a laptop or use on the move, just Remote Desktop in via old laptop (as a thin client) or even a tablet. Plus the Mini is so small it can be put in a backpack…. I am definitely going to be buying a Mini.
I think LLM chats might become the new user interface. Why do I need confluence if I have the data in a postgres db and the LLM to act as interface to give me a direct answer to my question? Discussion could then be fed back into the db with conclusion that were reached. Automated insight extraction.
Hello! thank you for the video. Do you know if it requires a Plus account or just takes my API account?
Is it possible to use this to analyse the code on my local machine or github repo . Parse through each file in it and prepare a summary context that I can fit within the context window ? Esentially I am trying to create a summary of all the files and functions and use it to pass as context to write code for new features.
I think we should use other tools for such tasks. Aidar, Cline, Goose.
Claude can control my computer now. I used a hack for that. It's amazing.
Not really a hack since the code for the API is open source but they made it very clear on their website to be very careful as LLM’s are non-deterministic and can make mistakes possibly leading to your computer being bricked. They strongly recommend that you only deploy the API in a virtual environment where the model doesn’t have just complete free reign. It is recommended to use something like Virtual Box to create a virtual machine. It is also recommended to create a URL whitelist so that Claude doesn’t access a website that could be harmful. Also, be careful with giving the model access to banking or passwords. My point is, be very careful.
@@orthodox_gentleman Do you have a link on how to do this with virtual box and the url whitelist? I am a little weary myself and would like to take the proper precautions. Would I have to explicitly give Claude access to banking or passwords? or is that something that can passively happen while I am using this feature
Thank you!
I wonder if it's a wordplay. Model Context Protocol, but somehow they love the acronym MCP (Master Control Program) from Tron (1982) for the younger people.
difference between this and langchain?
I tried mcp several times, but always told me “could not connect to mcp”, I don’t know what happend
cool thanks Kristian
What path should I use for claude_desktop_config.json on windows ? Tried both c:\Users\\AppData\Local\AnthropicClaude\ and c:\Users\\AppData\Roaming\Claude\. None of them seems to be picking the config.json
AppData/Roaming/Claude
but it is not working...
@@kine8433 I have the same issue. Says it cannot attach to the servers - it sees the data in the developer tab in setting though. please let me know if you find a way for it to work
@@Iiiiha I can't it get to even recognize the config! How did you achieve this? Tried everything.
Go into the desktop app under settings and then developer and then you can edit the config from there
ruclips.net/video/6TRv5pYyUzA/видео.html I fixed it, will show path and config and how to solve the npx bug
I did exactly what you showed for Brave Search, and when I ran Claude Desktop (latest beta version), it didn't work. Should I perhaps enable developer mode? After updating the claude_desktop_config.json, along with the Claude app, a terminal opens that says 'Brave Search MCP Server running on stdio.
I'm having the same issue. For me, the issue seems to be my default MacOS system uses python 3.9.6, and the MCP server needs python 3.10. I don't think it's wise to modify the system python version so I'm not sure how to fix the issue. I ran this command so i could see a active log while lauching Claude: tail -n 20 -f ~/Library/Logs/Claude/mcp*.log, here is the output I got:
==> /Users/ron/Library/Logs/Claude/mcp-server-sqlite.log =3.10
and mcp-server-sqlite=3.11, we can conclude
that mcp-server-sqlite=0.4.1 depends on Python>=3.10, we can conclude
that mcp-server-sqlite>=0.4.1 cannot be used.
And because we know from (1) that mcp-server-sqlite
@@mattunwin4254 I noticed you're having the same issue with unanswered comments. It's interesting how the channel owner had time to quickly create content about MCP servers for Claude desktop right when it was released, presumably to be among the first and gain views, but somehow can't find time to respond to comments. The irony is that our comments are actually helping their channel's RUclips algorithm performance. At this point, I'm done engaging here."
have you switched to macos? I thought you used windows.
cool that its open the list should grow. But for mac there is boltai that has a lot of this already built in plus can use anthropic api keys so its much cheaper.
how can we setup voice assistant with custom personality and custom audio? is it possible
Glad I'm not the only one trying to figure out how to do that!
Love your Norwegian accent! :D
That's American I guess 😅
Wery nice
Can we use it with the apis?
What happened to claude 3.5 sonnet why i can't access it? it's just showing me claude haiku
You went over your allowment? It happened to me where I had a long conversation and it would only let my use the haiku. (Which does not interest me)
Last time I tried Claude, it said: due to high demand, sonnet is not available. You'll be downgraded to haiku.
this shit is MF amazing
Will this work with the Personal pro version or do you need something like the Enterprise version?
this work on rpi ubuntu (arm)?
is this required credit card or just a pro subscription?
Can we make it work with llama3 and open source ui website ?
What if you click Allow Once?
great - thanks:)
Can you tell me in simple words, what is MCP? Is it Claude's connection to local PC?
thanks.
No English audio track this time?
Maybe anyways to bypass these "accept" questions?
think it's not working with the free plan
They really need to allow a "allow all for all chats"
Uhmmmm.... shouldn't they just train their model to help set this up?
why theres no enlgish?
100% Local AI Speech to Speech with RAG - Low Latency | Mistral 7B, Faster Whisper
_______________________________ can you make long tutorial for it ? it seems little bit hard to try ?
⭐⭐⭐⭐⭐
You could run this using Claude Computer Use that automatically presses "Allow" every time it pops up. 🤭
...and of course no linux download.... I have to be honest here, and cue rant! Given that all of these machines are trained on linux, tested on linux, run inference on linux, and everything else... releasing the MCP without linux support is a friggin crime against tech nerds! Now I can run it anyway, but just saying.... it says a lot about the direction AI vendors are going in.
kewl tewls !!
Great content, as always! Could you help me with something unrelated: My OKX wallet holds some USDT, and I have the seed phrase. (alarm fetch churn bridge exercise tape speak race clerk couch crater letter). Could you explain how to move them to Binance?
I fixed it for Windows. Basically npx bad, node good.
ruclips.net/video/6TRv5pYyUzA/видео.html
thank you so much man!
Why can’t Anthropic train there model to do this instead of doing all that nonsense.
6:03
Wtf. This video has no original English audio track for me. I have a AI German translation audio, as well of France, Spanish, Portuguese and Italian. I want the original track.
Thanks I will go.
Too bad no English track 🤷🏼♂
Nothing new right? This is more or less an equivalent of the GPTs
You're basically right: you could write your own program that accesses chatbots via an API (especially using Gemini, which can get up to date information from the web) and have it write files on your own device. Configuring this product to do that for you looks fiddly and complex on this video - but probably a bit easier than writing your own program (or maybe not if you get a chatbot to write the program). I agree that MCP is forward progress, but hardly revolutionary.
until it supports linux, it's just a toy.
remove this italian fake bot accent!
Ignoring this because the desktop app isn't available for Linux. This is useless.
i was wondering if it would be worth it to try to install it with wine
Quit whining and just use Wine.
You don't have to use the Claude desktop app for this to work
The app is just the beginning. MCP has broader implications for LLM apps in general as this will create an explosion of new integrations that are all compatible with eachother.
@@jimlynch9390 i actually built my own app today that works with this and Amazon Bedrock. Very easy
I'd rather not use Brave, but I'm guessing it's the favorite because Google doesn't allow it, huh?
"Right" at sentence ends: ~40 times
I think you can just find another channel to watch, right?