Works well! Tried with ollama sqlite. An interresting new cli option could be to transform it in "server" mode able to receive / respond to messages coming from api calls so the chat+mcp will be available from remote computers/agents
Amazing Chris! I see your video with Claude Desktop which seems perfect, but kind of blackbox form me. Now you change it to devevoper friendly playground with source code explanation - big thank you !!!
glad you like it, yeah the first video was a quick intro video that i released same day MCP came out, but i wanted to do my usual deep dive thing. and it just made sense to make it work with ollama at that point, and have a sensible cli that helps demystify it
@@chrishayuk I don't like "black boxes" but deep dive into source code takes huge effort for me usually (but no pain no gain :D ). Your channel (content/explanation/code examples) seems to be fit for my needs - good luck!
@@michalsestrienka4636 thank you, appreciate it, i try and always go one level below if i can, not always possible if i want to do something fast, but i will try and at least follow up if i think it's too shallow. thanks for enjoying the channel, and glad it's useful to you.
I've been wondering, anthropic have many optimizer such as in context rag, prompt caching, and some other more i couldn't remember. I wonder does this MCP server also optimized ready? or we have to optimize them by our self?
how does this compare to autogen, how can they work together? seems some overlapping but not really... can't quite wrap my head around them for integration
i will probs do a video on agents and mcp at some point.. but think of this as a way of giving structure to agents. so complimentary as opposed to replacing
That's awesome! Would it be possible to connect to other databases (eg. MSSQL) and call up stored procedures specificially, instead of relying on AI to write the SQL?
yep, it's possible, you might need to change the system prompt, and will need an mcp server with function calls that handles sql server but it will work
@@chrishayuk Thanks for that. I'm using the windows version of Ollama, and when running the main example, I get the following error: Error processing message: Ollama API Error: [WinError 10049] The requested address is not valid in its context Any ideas on whats the cause? Ta
This is what I expected from MCP, not having it tied to Claude Desktop, good job !
yep me too, hence why i did this video, and built the tool, glad you liked
@@chrishayukso basically you created an adapter that’s is an MCP host and spits out “tools” for “tool use” in the usual tool use way? Is that correct?
Works well! Tried with ollama sqlite. An interresting new cli option could be to transform it in "server" mode able to receive / respond to messages coming from api calls so the chat+mcp will be available from remote computers/agents
It’s like you’re reading my mind
This is exactly the video i was looking for. AI is crazy, every day news worth seeing
glad it's useful
Great job Chris in creating and explaining this!
thank you, it was a lot of working getting this video together, mainly building the tool
Legend. Just came across MCP and was hoping someone would build a client as a demo. Thanks.
welcome, glad it was useful
A wonderful implementation and explanation!
yaay, glad you liked :)
Amazing Chris! I see your video with Claude Desktop which seems perfect, but kind of blackbox form me. Now you change it to devevoper friendly playground with source code explanation - big thank you !!!
glad you like it, yeah the first video was a quick intro video that i released same day MCP came out, but i wanted to do my usual deep dive thing. and it just made sense to make it work with ollama at that point, and have a sensible cli that helps demystify it
@@chrishayuk I don't like "black boxes" but deep dive into source code takes huge effort for me usually (but no pain no gain :D ). Your channel (content/explanation/code examples) seems to be fit for my needs - good luck!
@@michalsestrienka4636 thank you, appreciate it, i try and always go one level below if i can, not always possible if i want to do something fast, but i will try and at least follow up if i think it's too shallow. thanks for enjoying the channel, and glad it's useful to you.
Interesting development, Chris.
thank you, glad you liked
I've been wondering, anthropic have many optimizer such as in context rag, prompt caching, and some other more i couldn't remember. I wonder does this MCP server also optimized ready? or we have to optimize them by our self?
great video! can you make a follow up on how to get this integrated into something like Open WebUI?
Amazing work :)
awesome, glad you liked
how does this compare to autogen, how can they work together? seems some overlapping but not really... can't quite wrap my head around them for integration
i will probs do a video on agents and mcp at some point.. but think of this as a way of giving structure to agents. so complimentary as opposed to replacing
Great video
Glad you enjoyed it
That's awesome! Would it be possible to connect to other databases (eg. MSSQL) and call up stored procedures specificially, instead of relying on AI to write the SQL?
yep, it's possible, you might need to change the system prompt, and will need an mcp server with function calls that handles sql server but it will work
@@chrishayuk Thanks for that. I'm using the windows version of Ollama, and when running the main example, I get the following error: Error processing message: Ollama API Error: [WinError 10049] The requested address is not valid in its context
Any ideas on whats the cause? Ta
Excellent job. I've cloned the repository but I don't find the requirements.txt file, how can I solve it?
I used uv rather than pip
Just install uv("pip install uv"), and then cd to the mcp-cli, then ("uv add requests"), Done
@@zhanpengli321 Thank you very much, it works.
good shout, i've updated dependencies and requirements
This would be cool to use as a host with FastMCP for servers
I do understand the olama obsession. Just use Claude, it’s already built and functional
DeepSeek would be the best Alternative to claude
, i think
good shout, not tried deepseek with it, qwen works really well
Nice