From REST to MCP (1/2): Different Dimensions

Short summary

This article explains why directly wrapping REST APIs as MCP tools is a bad default, comparing the two across several dimensions. In REST, developers encode control flow during development; in MCP, the AI agent constructs workflows at runtime, choosing tools and interpreting results. Key differences include who selects request parameters, how errors must be communicated (actionable feedback vs status codes), and how MCP tool results persist in conversation context unlike stateless REST responses.

• •REST encodes control in application logic; MCP shifts control to the AI agent at runtime
• •MCP tools need richer error handling—agents must interpret failures and decide next steps without pre-written client logic
• •MCP tool results persist in conversation history, unlike stateless REST responses, creating context-window cost implications

Generated with AI, which can make mistakes.