The Model Context Protocol (MCP), an open standard released by Anthropic in November 2024 to connect AI agents to external data sources, is facing criticism regarding its long-term necessity. While MCP was designed to bridge the gap between isolated models and fragmented data silos, the rapid advancement of Large Language Models (LLMs) is changing the technical landscape.
PolicyAnthropicModel Context Protocol
Criticism Arises Over MCP as LLMs Gain Ability to Call APIs Directly
Current discussions suggest that as models become more capable of reasoning and autonomous execution, they are increasingly able to interact with documented HTTP APIs and Command Line Interfaces (CLIs) directly. This capability potentially bypasses the need for specialized MCP servers that wrap these existing interfaces.
The rise of "context bloat"—where multiple MCP servers overwhelm a model's context with various schemas—has also prompted the development of alternative patterns. Tools like Composio, MintMCP, and Pipedream already offer methods to manage external service credentials and toolsets more efficiently. Furthermore, the growing adoption of standards like the Accept: text/markdown header allows agents to request human-readable text directly from documentation sites, providing a streamlined alternative to verbose JSON or XML responses.
As agents gain the ability to execute code and manage complex workflows with minimal intervention, the industry may see a shift from specialized protocols toward standardized direct communication with web services and operating system interfaces.
Sources
- Why MCP Was Always a Bad Idea? （Hacker News Frontpage, 2026-09-20）
- Anthropic公式発表