In November of 2024 everything changed, but it wouldn't be till January / February of 2025 that companies started all jumping on the MCP bandwagon of creating tools and "plugins" for AI's like Claude and Cursor. The official GitHub repo from Anthropic has now over 3400 commits since November 2024, and those are just people adding new MCP's to the list.
Now I am not going to get into what an MCP is, there are millions of YouTube videos and blogs about those, but the quick TLDR is that an MCP is basically a way for an AI like Claude (Not ChatGPT because it doesn't support MCP's yet...) to interact with real world programs and services. So a great example is an MCP for MariaDB which allows Claude to interact with your database, or an MCP for Figma which allows Claude to create stuff in Figma for you. (https://modelcontextprotocol.io/docs/getting-started/intro)
With the whole AI buzz going around, a huge amount of companies have hopped onto the MCP wagon so to speak, companies like Notion, Stripe, Twilio and far far more, but my questions is this, at what point does an MCP just create a huge amount of chaos for your app or software, I mean do you really need an MCP to talk to your database? People have been building database queries and commands forever with mysql and Sqlite (and more) so why now just everything goes AI?
It almost feels like we are going back in time, because the amount of compute power that goes into getting AI to run MCP commands could be halved if you just pass to it what you need... A good example is my ClearTask WhatsApp bot I built. I had the option of adding an MCP to pull data from my database for the notion tasks, but if I did go that way, I would need to place a whole bunch of AI commands in place to make sure the LLM didn't decide to pull another users data instead of the person querying the data. So instead, I just pass the users notion data to the LLM in order for it to have all the context it needs to fulfil the users request. No MCP's needed.
On the other hand, they can save time, when building something locally if I want to have a quick lookup of a sqlite database instead of having to open up terminal or a app to go through it, asking an AI todo can save time, but when it comes to Production deployment, I have thoughts...
How do you setup your software stacks? I would be interested if you incorporate MCP's into a production app and what that MCP does!