Why MCP Is Mostly Bullshit
a year ago
- #Hype
- #AI
- #MCP
- MCP (Model Context Protocol) is a standard for building tools that provide additional context to AI assistants, launched by Anthropic in November 2024.
- MCP aims to replace libraries like LangChain and LlamaIndex by allowing custom integrations as tools accessible to AI assistants via APIs.
- The protocol generalizes tool use, enabling various services (e.g., vector databases, filesystems) to be queried by LLMs as tools.
- An example MCP server implementation for the Qdrant vector database demonstrates how tools can be decoupled from LLM applications.
- MCP is not the only solution for tool decoupling (alternatives include LangChain, LangGraph, SmolAgents, LlamaIndex), but it has gained sudden popularity due to hype and architectural appeal.
- Criticism includes hype-driven adoption and the tendency of software engineers to favor architecture over problem-solving ('architecture astronauts').
- MCP is seen as a useful but not essential approach for enhancing AI assistant context, amid broader skepticism about trend-driven development.