MCP
Model Context Protocol
Adding here all that I am learning about Model Context Protocol.
MCP is released by Anthropic in 2024
It is an open standard enabling seamless integration between an LLM and any other external application or data source.
MCP Architecture
Image Source: https://modelcontextprotocol.io/introduction
LLMs can now actually do the work like raising a Pull request, emailing, updating something in database etc.
Developers can either expose their data through MCP servers or build AI applications (MCP clients) that connect to these servers.
An MCP server is responsible for feeding the context to the LLM Client, even though it can get more data or info from outside data source, it will only ingest the data that the client needs.
There are many kinds of contexts that we can give to a MCP server for example - Resources, Prompts, Tools, Sampling, etc.
MCP Server works on STDIO (Standard input/output)
Communication between a client and server is handled by a Transport. Two built-in transports are
STDIO - standard input output - for local MCP servers and command line tools.
SSE - Server sent events - when MCP server is hosted externally, we can use POST requests.
In future we can see different companies coming up with their own MCP servers, and we can seamlessly integrate that to the LLM/Host to get the desired result that we want.
Sources/References: