According to @godofprompt, Anthropic has introduced code execution with MCP, addressing one of AI’s biggest bottlenecks—token inefficiency in agent operations (source: Twitter, Nov 6, 2025). Previously, agents would use extensive tokens for every tool call, definition, and intermediate result, leading to context overload and increased risk of data leakage. With MCP, agents now write code to call tools directly, reducing token usage by 98.7% and completing tasks up to 10 times faster. This approach, also referred to as 'Code Mode' by Cloudflare, eliminates context overload and minimizes data leakage, signaling a major shift in AI agent architecture. The business impact is substantial: organizations can deploy more efficient, scalable AI agents with lower operational costs, opening new opportunities in process automation and intelligent workflow optimization.