MCP Server Could Have Been a JSON File (materializedview.io)

The Model Context Protocol (MCP) server, designed to enable large language models (LLMs) to interact with external software by defining available tools and resources, is facing skepticism over its necessity. While MCP aims to teach LLMs how to call APIs or CLI tools to affect real-world actions, critics argue that its architecture—using static JSON definitions for prompts, resources, and tools—overcomplicates what could be achieved with existing standards like OpenAPI or traditional CLIs. For example, MCP’s "searchFlights" tool definition closely resembles an RPC schema that OpenAPI already handles effectively, and LLMs like ChatGPT can seamlessly interpret OpenAPI specs and complex CLI commands. The significance for the AI/ML community lies in questioning whether MCP is a needed innovation or a temporary workaround. Although MCP may offer gains by condensing tool definitions into a lighter format that fits limited context windows, the rapid expansion of model capabilities to multi-million token contexts diminishes this advantage. Moreover, well-documented APIs and powerful CLI interactions already provide robust, widely adopted methods for LLMs to interface with external systems. Most MCP adoption appears confined to poorly documented enterprise services, where arguably the root problem is inadequate API standards rather than a new protocol. This debate highlights a broader takeaway: advancing AI-agent productivity requires building better developer tools and interfaces rather than inventing new protocols that essentially replicate existing functionalities.