I have not yet been too much impressed with all the AI hype so far. But two weeks ago, @dsebastien mentioned to me in a PKM session the concept of #MCP and oh boy, that is mind blowing!
I configured both Todoist and Playwright MCP servers and asked Claude to find all my todos with links to Instagram (reels), extract the metadata for each post and add that metadata to each todo as a comment.
And it just worked!
@schmkr This sounds super interesting, wanna describe the UX a bit more? I'm kinda looking from the outside on that tech atm and trying to figure out what the UX is generally like and what it can provide for you
@schmkr Ofc feel free to just send me to RTFM as well
@alkoclick the ui is still a chat interface. But what it does is give the LLM more abilities to perform specific tasks on the machine, via the Claude desktop app. It spawns separate processes for each client, which can be a NPM package, python repo or a container image. At least those are the examples, I can imagine in theory it can be anything as long as the spec is followed. Tomorrow I can share some screenshots of this particular example of mine. I left my laptop downstairs and am off to .
@schmkr Yep, it's super cool. And now that OpenAI has embraced it, it has become an industry standard.
Both Zapier and Cloudflare have added support, and so this is going to become a huge market.
Security remains a challenge, but things are getting better.
And the list of MCP servers grows dramatically fast.
In module 8 of my upcoming knowledge management course, I'm covering MCP as well as many other AI tools. Can't wait to publish all those videos!