Awesome.. I think even just based on your simple example, being able to improve on what Claude desktop can do and personalize it with my own data sounds exciting
The device you're using is already listening and collecting data on you constantly. Same with every app you use and they sell it to whoever wants it already. We have no security in this world.
It's an excellent video. Thanks very much. I've connected Claud to my own db and start asking him about it. Sometimes, it just answered with non existing columns or misunderstood the meaning of some of them. How can I pass him the full description of the tables?
it's basically the system prompt, you can convince claude to go read the tables before attempting to select the data... that was the main trick when i built my version
there is nothing stopping you reading the tables at the beggining of the conversation, and the stuffing that into the context. a more sophisticated approach might be to raq the query, and pull the best system prompt to cover it
I really appreciate your patience. In my case, I have more than 100 tables. I am afraid it won't be a good practice to give their description as a system prompt. Should I use RAG or any other technique for my case?
openweb ui can't intereact directly with your database though, this is a game changer because companies with massive data bases with hundreds of tables and billions of rows can now have their databases interact directly with the AI. this will make it much easier for example to just build a dashboard for whatever data you want to track using sonnet. the Model Context protocol has literally murdered so many startups that were charging to do this for people.
@@Aurelnpounengong _"openweb ui can't intereact directly with your database though"_ This is just wrong. It can interact with whatever it has a tool for.
I'm trying but I get the following error I'm not a professional VSC I'm probably making a mistake somewhere but I don't know where =) Could not load app settings There was an error reading or parsing claude_desktop_configjson: SyntaxError: Bad escaped' character in JSON at position 180 (line 8 column 15)
@@chrishayuk Awesome, looking forward to it. It would be great if you also cover, or gave your opinion in the comment :) how this compares to RAG. What is the use case for MCP and what is the use case for RAG? Seem to do similar things as far as I can see, specifically when it comes to connecting own data and chatting with it.
I did not hear anything about authentication, encryption and cybersecurity guidelines.................I would wait until you get a proper safety review and several use cases before jumping in. My 2 cents worth.
Wish there was an example to build clients. Right now using the Claude Desktop it is easy to run out of sessions before you are asked to pay for a subscription. I just wan to play with the protocol first before I spend $$ trying to explore.
@@seanwfindley very much appreciate it and glad you love the channel and find it useful, THANK YOU!!!, but honestly everyone has different styles, and not everyone is gonna like my channel or style,. i'm super cool with that
This is the first time I have ever watched this guys channel. Your comment is idiotic and that makes you a 100% IDIOT. What an exceptional piece of content, I compliment you. I did not know about MCP until this morning. Tonight I connected to an external postgres database. Nice work my friend.
Great explanation of what we have built. I much appreciate the view and that everyone is just as excited as we are about it.
@davidsoriaparra8833 you've done an amazing job here, i see a tonne of innovation coming, and this is the big opener for federated ai powered services
very cool how you did the code overlayed onto the video - well done
Anthropic is full steaming ahead. Beautiful to see this API stuff coming into being so fast.
Everyone I know is hyped for this.
Awesome.. I think even just based on your simple example, being able to improve on what Claude desktop can do and personalize it with my own data sounds exciting
Best explanation of MCP so far!
Wow, thanks!, glad it was useful, hope you enjoy my new deep dive video also
This has considerable security implications. Will need to look at the docs.
yeah, there is definitely gonna be some fun with this one
as usual. custom gpts are really cool to exfiltrate data from openai
The device you're using is already listening and collecting data on you constantly. Same with every app you use and they sell it to whoever wants it already. We have no security in this world.
great overview, thanks.
subscribed
thaaaaaank you, and glad you liked
I thought Dario had made a video about this himself! 😊 Haha. Good video! Thanks I had no idea what this was.
glad it was useful
Keep us informed when you come across ways to integrate this with local LLM agents like CrewAI or Kotaemon
that is a good shout, i really wanna do some further deep dives with this, especially with agents
How does Claude Desktop know when to pick a tool and which one? How long can be the list of installed tools.
It's an excellent video.
Thanks very much.
I've connected Claud to my own db and start asking him about it.
Sometimes, it just answered with non existing columns or misunderstood the meaning of some of them.
How can I pass him the full description of the tables?
it's basically the system prompt, you can convince claude to go read the tables before attempting to select the data... that was the main trick when i built my version
Ok.
And can I do it only when I start my application, or I have to do it for each conversation?
there is nothing stopping you reading the tables at the beggining of the conversation, and the stuffing that into the context. a more sophisticated approach might be to raq the query, and pull the best system prompt to cover it
I really appreciate your patience.
In my case, I have more than 100 tables.
I am afraid it won't be a good practice to give their description as a system prompt.
Should I use RAG or any other technique for my case?
How is it different from tools for example in open-webui?
openweb ui can't intereact directly with your database though, this is a game changer because companies with massive data bases with hundreds of tables and billions of rows can now have their databases interact directly with the AI. this will make it much easier for example to just build a dashboard for whatever data you want to track using sonnet. the Model Context protocol has literally murdered so many startups that were charging to do this for people.
@@Aurelnpounengong I think the similarity is more with Open Iterpreter @tooluse
@@Aurelnpounengong _"openweb ui can't intereact directly with your database though"_
This is just wrong. It can interact with whatever it has a tool for.
Good stuff!
Glad you enjoyed it
nice lights its like i am in thailand
In laymans understanding - does this need claude api or works via the $20 subscription?
20 bucks subscription
I'm trying but I get the following error I'm not a professional VSC I'm probably making a mistake somewhere but I don't know where =)
Could not load app settings
There was an error reading or parsing claude_desktop_configjson: SyntaxError: Bad escaped' character in JSON at position 180 (line 8 column 15)
Anyone know if one can use this with a custom developed client? i.e a native app, that has a chat bot? i.e I want to use my own app, not Claude.
And that video is coming today or tomorrow
github repo for the example is here: github.com/chrishayuk/mcp-cli
@@chrishayuk Awesome, looking forward to it. It would be great if you also cover, or gave your opinion in the comment :) how this compares to RAG. What is the use case for MCP and what is the use case for RAG? Seem to do similar things as far as I can see, specifically when it comes to connecting own data and chatting with it.
{}
Error executing code: MCP error -2: Request timed out
i am getting this error. can u help me figure some solution?
How secure is the data that Claude reads, does Anthropic confirm that the data remains on my server?
It can't by definition. If you are passing data to an llm to work with that data has to be sent to the llm server to be processed.
I did not hear anything about authentication, encryption and cybersecurity guidelines.................I would wait until you get a proper safety review and several use cases before jumping in. My 2 cents worth.
This is why I talk about the use cases being more in the cloud rather than local
@@chrishayuk Good Point!
So no more RAG?
Isn't your channel bad for viewer eyes as it has lot of blue light. It's cool.. but too cool that's it 🔵🔵🔵🔵
hahaha, yeah, maybe wear blue blocker glasses like i do
Wish there was an example to build clients. Right now using the Claude Desktop it is easy to run out of sessions before you are asked to pay for a subscription. I just wan to play with the protocol first before I spend $$ trying to explore.
video goes out either later today or tomorrow on how to do this... but here's my github repo with the example. github.com/chrishayuk/mcp-cli
So. Blue.
The guy is wearing blue light blocking glasses and fills the room with blue light...
You're evil for the fact that you beam blue light to your viewers and yet use a convenient blue light filter for your own wellbeing. Aren't you?
caught
Too much waffle sorry, stopped watching it.
yeah, i'm definitely not the channel for you, but thanks for trying
@@chrishayuk Screw those guys. Love your channel. Just enough information to get the rest of us going.
@@seanwfindley very much appreciate it and glad you love the channel and find it useful, THANK YOU!!!, but honestly everyone has different styles, and not everyone is gonna like my channel or style,. i'm super cool with that
This is the first time I have ever watched this guys channel. Your comment is idiotic and that makes you a 100% IDIOT.
What an exceptional piece of content, I compliment you. I did not know about MCP until this morning. Tonight I connected to an external postgres database. Nice work my friend.
🥱