Another helpful one. I really like how your videos have some practical applications and not just hype. I wish I could use the desktop app via a local api, that way you could make a separate app that can call claude with some MCP capability.
Does claude itself not adhere to MCP? I feel like it should be possible, maybe not through the desktop app but through their API? Idk if their API gives access to MCP servers though
This is the natural progress of tech. Same with virtual machines, when we started orchestrating them with Docker etc. The power of AI is in this type of setup, let each LLM be very specific to a subject and combine them like this to get a perfect end result. I'm just waiting for this to become more stable and produce consistent high quality.
I love what you're doing with these protocols and how you already understand how to put them together. I would be curious to see you do the River Problem as a single prompt.
I have been using Windsurf with Claude and have learned that anytime it gets stuck it is worlds better to give the situation and the code in question to o-1 preview and then give that response back to Windsurf. I have been pretty amazed at that combo.
L@@penrewurst3527 o1 solves the problem > you explain the problem to windsurf and give it the code. Windsurf is too stupid currently to do any problem solving, it was good 2 weeks ago now it's not even using clayde, theyre using some small random model thats horrible
The integration with TH-cam to generate thumbnails is a smart move! It’s a reminder of how AI can streamline even the smallest aspects of content creation.
What an awesome video, will this setup also work with the new o1 full model when API is released? Between the coding capabilities of both, it may make the $200/mo plan worth it
it's sort of like agents, based in claude desktop, you run o1 for design, architecture, deep think stuff, then let claude sonnet fill the details, and the code. it might work well, or perhaps o1 as spec writer... nice experiment
It would be great that you create a client also that works with a server, instead of using the Claude Desktop app! But great example, of the best so far in the internet!
Maybe the result wasnt what u hoped for. But the possibilities are infinite. Maybe specificify in the prompt a specific ChatGPT oriented for graphical design. This Will give more creative prompts/designs.
How well does this handle longer code? I’ve had issues with anything over 300 lines. It has to be broken into portions. Does this handle that on its own?
Any idea how to make it work with PDFs? I tried setting up a file system server and it works great with code repos, but it can't read PDF files (they have to be uploaded to the chat or via API). Would there be a way to actually do that? Would be super useful.
It’s special, but the thing is it’s still the one-eyed leading the blind because these models are not truly intelligent, so one model is liable to replace mistakes with mistakes of a different kind. But I am tinkering and building gadgets too, because it’s fun and it might land at some point in the future.
Another helpful one. I really like how your videos have some practical applications and not just hype. I wish I could use the desktop app via a local api, that way you could make a separate app that can call claude with some MCP capability.
Damn fellow legendary account I remember you
Does claude itself not adhere to MCP? I feel like it should be possible, maybe not through the desktop app but through their API? Idk if their API gives access to MCP servers though
This is the natural progress of tech. Same with virtual machines, when we started orchestrating them with Docker etc. The power of AI is in this type of setup, let each LLM be very specific to a subject and combine them like this to get a perfect end result. I'm just waiting for this to become more stable and produce consistent high quality.
I love what you're doing with these protocols and how you already understand how to put them together. I would be curious to see you do the River Problem as a single prompt.
Impressive that it was able to combine a few tools in one response. That gives me hope, as I thought I'd need a kind of orchestrator in my server.
I have been using Windsurf with Claude and have learned that anytime it gets stuck it is worlds better to give the situation and the code in question to o-1 preview and then give that response back to Windsurf. I have been pretty amazed at that combo.
How do you mean ? So you just copy paste the code and explain the problem to o1 and then using that answer as a advice for Claude ?
L@@penrewurst3527 o1 solves the problem > you explain the problem to windsurf and give it the code. Windsurf is too stupid currently to do any problem solving, it was good 2 weeks ago now it's not even using clayde, theyre using some small random model thats horrible
The integration with TH-cam to generate thumbnails is a smart move! It’s a reminder of how AI can streamline even the smallest aspects of content creation.
What an awesome video, will this setup also work with the new o1 full model when API is released? Between the coding capabilities of both, it may make the $200/mo plan worth it
it's sort of like agents, based in claude desktop, you run o1 for design, architecture, deep think stuff, then let claude sonnet fill the details, and the code. it might work well, or perhaps o1 as spec writer...
nice experiment
This something meaningful about AI agents now .
It would be great that you create a client also that works with a server, instead of using the Claude Desktop app! But great example, of the best so far in the internet!
Maybe the result wasnt what u hoped for. But the possibilities are infinite.
Maybe specificify in the prompt a specific ChatGPT oriented for graphical design. This Will give more creative prompts/designs.
What did run you in tokens and USD?
We just need a built-in tool for executing code in the chat UI!
if i understand correctly i could also easily connect my local ollama (or similar) with an openai compatible api to claude as well? niiiiceee
How well does this handle longer code? I’ve had issues with anything over 300 lines. It has to be broken into portions. Does this handle that on its own?
what the thumbnail pic? Are u hosting a talk?
Any idea how to make it work with PDFs? I tried setting up a file system server and it works great with code repos, but it can't read PDF files (they have to be uploaded to the chat or via API). Would there be a way to actually do that? Would be super useful.
thanks:)
Great
Looks like a lot of code for what amounts to a constraint satisfaction problem.
So, how many people are prepping your videos, coding, testing, mlops, etc?
I add the tools as per the documentation but they don't get detected in the desktop app. Anyone else have this issue?
use chatGPT to tell you whats wrong or claude web interface, what i did with my issues.
do a mongodb MCP
Do you need Claude subscription for this ? Or do you need Claude api credits for this ? In addition to open api key :( seems expensive
Works with the free version of Claude as well.
The challenge is not making more MCP servers but a new host to replace Cloud Desktop.
what do you mean explain a little please.
Nobody want to see my movies?
It’s special, but the thing is it’s still the one-eyed leading the blind because these models are not truly intelligent, so one model is liable to replace mistakes with mistakes of a different kind.
But I am tinkering and building gadgets too, because it’s fun and it might land at some point in the future.
The audio is auto translating on mobile, I don't want to hear you in my native language, the translation sucks, unsubscribed
That would be youtube who did that, and not him. And you can switch it back to English in the settings