Evan Swanson, who manages brands doing $60M a year, keeps an "operator OS" folder with his operating principles and brand guidelines and opens it in Claude Code or ChatGPT Codex, so he can switch AIs without losing context; the folder is backed up to GitHub. He connects MCPs instead of all-in-one tools, including BrandSearch (competitor traffic, landing pages, social and Meta ads), Higgsfield (image and video generation), Shopify, Canva, Gmail and Slack, plus Amazon research and PPC tools. One of his prompts: use the BrandSearch MCP to scan competitor websites for their top-performing ads and landing pages, then use the brand guidelines in the folder and Higgsfield to generate similar static ads. Others review PPC performance against a 20% margin target and estimate cost per acquisition from keyword CPC and conversion rates.
Prompt
Use the brand search MCP to scan competitor websites for their top-performing ads and landing pages. Then use my brand guidelines and Higgsfield to generate similar static ads according to our guidelines in the folder