I missed posting for the last couple of days as I was busy with a family function.
On the learning front, I've been exploring MCP (Model Context Protocol). Learning AI-related technologies not only improves productivity but also becomes a valuable addition during job hunting and technical interviews.
Understanding MCP has been on my learning list for quite some time, and I'm finally spending time learning it properly. My goal is to build a custom MCP server, and I'll definitely share the GitHub repository once I have something useful to demonstrate.
MCP (Model Context Protocol) is an open protocol that standardizes how AI applications communicate with external systems, tools, databases, APIs, and SaaS platforms. Instead of every AI application building custom integrations for each service, MCP provides a common interface for these interactions.
A typical flow looks like this:
MCP Client (Claude Desktop, ChatGPT, Cursor, VS Code, or a custom AI application)
⬇️
MCP Server (provided by an external service or built as your own lightweight microservice)
⬇️
External APIs, databases, SaaS platforms, filesystems, or internal enterprise systems
For example, Claude Desktop acts as an MCP Client and can connect to different MCP Servers, allowing it to interact with external tools through a standardized protocol.
An MCP server can expose three primary capabilities:
MCP communication is based on JSON-RPC 2.0.
If you're familiar with RPC (Remote Procedure Call), JSON-RPC follows the same concept but uses JSON as the message format. A request typically contains fields such as jsonrpc, method, params, and id, where the method identifies the operation to execute on the MCP server.
There's still much more to explore, including transports, client-server lifecycle, authentication, sampling, and building production-ready MCP servers.
More to come... Stay tuned! 🚀