Anthropic has open-sourced a revolutionary new protocol—MCP (Model Context Protocol)—which aims to completely address the pain points of connecting data in LLM applications! Its goal is to enable cutting-edge models to generate better and more relevant responses. No longer will you need to write custom integration code for each data source; MCP handles it all with a single protocol!
The Model Context Protocol (MCP) is an open protocol that enables seamless integration between LLM applications and external data sources and tools. Whether you’re building an AI-powered IDE, enhancing a chat interface, or creating custom AI workflows, MCP provides a standardized way to connect LLMs with the context they need.
The way and principles of how MCP works have been clearly explained in the official documentation, and anyone with a bit of computer knowledge can understand it.
-
MCP Hosts: Programs like Claude Desktop, IDEs, or AI tools that want to access data through MCP
-
MCP Clients: Protocol clients that maintain 1:1 connections with servers
-
MCP Servers: Lightweight programs that each expose specific capabilities through the standardized Model Context Protocol
-
Local Data Sources: Your computer’s files, databases, and services that MCP servers can securely access
-
Remote Services: External systems available over the internet (e.g., through APIs) that MCP servers can connect to
The officially defined MCP Servers are somewhat similar to a Sidecar pattern; it can be placed anywhere without issue, and the core connection method is this MCP.
I am particularly interested in:
Protocol First, Specification Driven
This protocol may not be the best or most optimal, but with Anthropic’s influence, the industry will surely catch up quickly. In fact, langchain and spring ai have already launched quickly; like this large diagram from spring:
The official list of servers already includes nearly a hundred, including familiar ones like DB, MySQL, SQLite; Git, FileSystem, Google Server, etc.
MCP is not a silver bullet; we cannot expect it to solve all problems. Large model “extensions” like RAG, Function Call, Tools, and Web Search are already effective and useful;
The challenge lies in user intent recognition. When users ask random questions, how can we find matching extension services? Imagine millions of GPTs; MCP is just a protocol and clearly cannot solve this, just as the HTTP protocol cannot resolve your company’s management process issues. Claude currently uses local configuration (claude_desktop_config.json) and manually selects plugins or roles (equivalent to ‘@’ or ‘/’).
A major highlight of MCP is that it establishes a standard and relatively complete protocol, which has significant guiding meaning for the collaboration between large models and application ecosystems. Imagine the value and significance of Open API for communication between various services. The challenge is how many people are willing to adopt the MCP protocol?
Referring to the style of OpenAI’s API, it still becomes the design standard for this industry.
It’s worth following up….
-
https://mp.weixin.qq.com/s/k6t9SWzCGLRhWOfLmWjmOw Anthropic Releases Groundbreaking Open Source Model Context Protocol (MCP)
-
https://mp.weixin.qq.com/s/blITLCReqoJKCogLc7kbfg What Is the New MCP Protocol by Anthropic? (Comic)
-
https://www.anthropic.com/news/model-context-protocol
-
https://modelcontextprotocol.io/llms-full.txt
-
https://github.com/modelcontextprotocol
-
https://modelcontextprotocol.io/quickstart#need-to-install-the-prerequisites
-
https://modelcontextprotocol.io/introduction
-
https://github.com/modelcontextprotocol/servers
-
ApiHug101-003 Four Principles
-
https://learn.microsoft.com/zh-cn/visualstudio/extensibility/language-server-protocol?view=vs-2022
-
AI Programming Assistant that Integrates Learning and Practice – ApiSmart
