Видео

I’ve been having this issue constantly. Also hitting the chat session wall that locked the chat session down. How I found your videos is I was looking for ways to use a different host for MCP instead of Claude desktop. My hopes was to find someone who has already achieved it with ChatGPT or Gemini.

I've used this tool. Or rather I told Claude to use it and it did to great effect!

This has been driving me crazy - thank you for the fix!

fun trip 😂

Hey Pat, hope you're having a good new year. Just wanted to say thanks for all these videos, they've actually provided real life use cases - which are actually useful. From the lightrag database to the google suite implementation, this will greatly increase productivity. I still haven't found a clear and visually appealing yet functional solution across all of youtube. I figured I'd ask you for your help directly as you are a professional in this and kind hearted. Any possibility of making a GUI (which can work on both macos and android/ios) which can essentially ~ do any productivity associated task for you? Without requiring continual approvals via clicking. And leaving it up to the user in what scenarios they want to provide confirmation before accepting . e.g Driving to work but still want to use a large amount of your brain power. In the shower but still want to use brain power. Walking to the shops. When you're just sitting around. Just ot name a few. ---> Simply Ask Assistant with your voice! e.g research all the drugs implicated in disease process X and provided sources such as RCT's from Pubmed. ---> Create a clean powerpoint presentation which captures the core information of this research. ---> Email someone (using my predefined emails to learn how i structure my responses) ---> Open X app and tell me what you see. Interact with the app. Click this button. etc. ---> Navigate to websites. ---> Auto arrange my to do list based on previous examples whats important to me. But read out the list aloud to me so i can verify. ---> Login to a website by tapping the fill button using bitwarden browser extenson. Just some stuff like this essentially. All of this in a nice GUI and with voice mode like advanced voice mode with chatgpt. (excpet with the ability to turn it off with voice)

Hi Pat- first off thank you so much you saved my day!!! also happy new year. One issue i ran into while using your code is it failed to install github. Also during installation, it says "No.env file found, using default configuration" and it proceeded to ask me to manually input GIT_PAT_TOKEN (or press Enter to skip). I input my Github token and pressed enter, not sure it if took it in. The key is also in the .env file that saved in the same directory that i saved the mcp_setup.py file... am i doing anything wrong here? Any insight you might have for me to configure GitHub? Any help is appreciated!

Using a middleware MCP server like you did here, can be generalized to make any other MCP server/service remote, right? Theoretically even lightrag could be running in the cloud, on a compute node somewhere (or another host in your LAN ;) ). Or, since lightrag uses its own vector storage, I guess theoretically that part could be decoupled and offloaded to Qdrant or other vector embeddings service (I would favor pgvector as I love postgres).. But yeah that would take some work.

Am I missing something here, I thought lightrag and qdrant are quite different RAG approaches, so not sure what you meant by "migration" from one to the other. I understand that Qdrant can be used for high-volume document queries using the usual vector embeddings similarity approaches (albeit multi-faceted, I got it). However only lightrag offers a higher conceptual/semantic level through the knowledge graph - and that part cannot be "migrated" to Qdrant.

Thank you for these updates Pat, they are great! I really appreciate your down-to-earth takes and especially sharing your work - that github repo is evolving nicely, I'm watching it closely. It'd be interesting to see if lightrag and quadrant RAG can be integrated to maybe complement each other in some cases?

Neat work! I'm wondering if there's a video about how to add additional MCP's to it? I'm wondering if there's one for Slack. I'm also wondering about the license--legally, I'm not sure if anyone can legally use it without a license?

Aliens are real, though😂

Thanks for your work 😊

It's Christmas time, great work, thank you

Nice one, Pat. Most of the MCP implementations are in Python. So it's good to see Typescript implementation.

love u man

You breal all with your exhibitionnism. I block you, what a miss!

Your wife will love all the email she will get because of this video.

I will try it too, Claude Desktop MCP will help me. It`s a fantastic tool.

are you reading my mind? was looking for kg RAG, and more lightrag hands on! Can't wait to see the code, if you are willing to share, as rough as it may be. Many thanks!

brilliant, thank you for sharing this, was looking for something like this myself

Thank you, I’m just not sure how to allow it access to local files?

Hey Patt I think you missed the part to mention that we need to navigate to google cloud console and then get the keys as json format and keep it in the local system to be able to use gmail services. Please correct me if I am wrong.

Nice work Patt. Hope to include more tools in future!

See later video I got email working with Ollama. It obeys.

Hello can you make it for me I try a lot but its not running

Awesome

I've been using it for a couple days and wouldn't say its crap at coding, it does give over complicated solutions to tasks most of the time but most of this can be solved with just giving it more detailed instructions on what you want and/or documentation. The benefit of spending $200 is you get unlimited usage so you can keep iterating on your prompt or give it smaller tasks that it puts more effort into. Obviously though for $200 you'd want it to give you a solution first time but even if it doesn't you don't have to worry about your prompt limit like with Claude. Glad you got your money back and your solution in the end up anyways.

I came here for t*it*s. But seriously... Clade 3.5 Sonnet after the last update also got quite stupid...

I was interested till i saw your presentation, fastest "Don't recommend channel" i have ever clicked battyboy

You mixed your onlyfans account with your YT acc.

OpenAi will took your job, your money, and even your shirt !

Thank you for sharing , keep it up , nice work

You gay?

imagine how many shirts you could've bought with that $200!

How would you go about creating Memory knowledgebase from pdf documents etc? For example Knowledgebase of big project with architecture documents, technology documents, code snippets and so on. Would you have to get Claude to generate all nodes and double check everything? Is mcp Memory even the right tool for that?

Update: I fixed it, will post soon

I've been experiencing the exact same limitations! I've been using Claude to help me develop new MCP servers for various applications and APIs. While the generated code is excellent and solves my problems so well, I have to specifically tell it to break down the implementation into smaller chunks so it doesn't generate too much at once. I need to request one code block at a time, and even then it still stop working many times. This forces me to start new conversations and re-explain the context, which is pretty annoying. Even just generating a single long TypeScript file for a MCP server becomes an issue as Claude gives up mid-generation. By the way, thanks for your video on fixing MCP filesystem issue on Windows, it saved the day.

Wasn't working for me until I killed the Claude processes in task manager. Otherwise Claude isn't truly restarting and it won't look at that config.

Hey Pat nice work! I has hoping you can shed some light on using local models with MCP instead of Claude.

Thanks a lot! I only fixed the file-system path, but now, multiple of my npm extensions work. Any idea why?

could you share us the mcp config desktop file i'm still struggling with github and search brave what doesn't work on windows

dude this worked for me. thank you

Your camera screen is blocking the code Patt!

Here's the repo with the python script. I'll update this repo as I add more tools to Claude on Windows. github.com/patruff/claude-mcp-setup/blob/main/README.md

Whatever you are doing on Github is not visible to us as Claude desktop is blocking the view so please kindly consider minimizing it next time.

You are very clever

You are king my friend

Your solution saved the day - thank you so much!

Dude you are like the only person on the internet that figured this out.

Not working for me. I don't have admin rights on my machine, could that be an issue? here's my claude_desktop_config.json { "mcpServers": { "filesystem": { "command": "C:/Users/<username>/nodejs/node-v22.11.0-win-x64/node-v22.11.0-win-x64/node.exe", "args": [ "-y", "C:/Users/<username>/nodejs/node-v22.11.0-win-x64/node-v22.11.0-win-x64/node_modules/@modelcontextprotocol/server-filesystem/dist/index.js", "D:/Data" ] } } }