Why MCP Was Always a Bad Idea

The Model Context Protocol (MCP), released by Anthropic in November 2024, was a protocol designed to help agents connect to external services and data sources. As LLMs became more advanced, the need for MCP started to diminish. With the growth in LLM adoption and improved models, agents were able to access external services directly through HTTP APIs and Command Line Interfaces (CLIs), making MCP servers unnecessary.

The LLMs became capable of writing scripts, reasoning about large codebases, and calling APIs directly, reducing the need for MCP. The Cloudflare Code Mode even demonstrated a more efficient way to use MCP by having LLMs compose calls into scripts. Most remote-service MCP servers were eventually deleted as agents with terminal access could replace them.

While there are still minor issues like CLIs returning verbose responses, the alternative already exists through documented HTTP APIs, standard content negotiation, and mature authentication mechanisms. The future of agents is now centered around using HTTP APIs and CLIs directly, making MCP a protocol of a bygone era.