For the past couple of months, the concept of MCP has been booming. MCP (Model Context Protocol) was developed and released by the end of 2024 by A...
For further actions, you may consider blocking this person and/or reporting abuse
I think MCP is great if you are integrating it in an ai platform like cursor and Claude which can by default use nlp to create actual db queries. However I tried to do a setup for integrating it into my own app which has my own llm models and whatnot. The project is completely from scratch so no Claude or cursor. What I found out that’s it’s just some flexible structure behind the scenes. I wrote my own mcp server for Postgres and it sort of felt redundant. Maybe I did not understand mcp clearly but if i am to create mcp endpoints and then call those endpoints according to user query which by the way I would have to generate by nlp. What is the point of mcp then ? If anyone could explain it to, please do, I will highly appreciate it. According to my understanding though, it might be good for already built ai agents like Claude and cursor but not for a completely custom product where agent is just a part of a project.
Hey, I had thoughts similar to yours. At first glance, it was just another way to wrap your existing APIs and endpoints.
However, when I dug a bit deeper, I found that MCP allows you to expose actions and resources to the agents in a way that makes them aware of what they can work with.
For example, if you want your agent to work with your Grafana instance and create dashboards, the agent would need to call the Grafana API. But in this case, it has no way to know what the API looks like, what endpoints to call, and what to expect. Through an MCP server, the agent knows what actions it can call and what inputs/outputs are expected (this is sorted through how MCPs are designed to have special endpoints telling the agent what it can use). I think of it as an interface that a service provider (Grafana in the example) can implement to allow agents to interact with it.
In its implementation, it is using HTTP (if we are talking about SSE servers), so it might seem like a simple proxy, but it's more about conforming APIs to a standard that agents can communicate with
I agree, in the end I think it’s about standardisation for ai agents.
MCPs are here to stay, Google is also doubling down on it!
Yes!
yeah this is awesome tbh, making all that setup easier is huge for me - you think more teams will switch to this kinda shared approach or do people hang on to old ways outta habit?
The old ways are not that old at this moment still 😅
We shall see!
💪
💪
👀
👀👀
Cool!
🤗
Cool article!
Thanks!
Great context example
🙏
Pretty cool feature, thanks for sharing :D
My pleasure!