The five-tool problem
Most teams use several AI tools at once: one for text, one for code, one for images and design. None of them knows what the others know. Each tool has to be told the brand, the project and the status again, and you are the one doing it.
Way 1: one shared document
A shared document with the brand, the tone and the rules beats nothing. But it has to be pasted in by hand, it gets long, and nobody knows whether the copy in the tool is the latest. It works until the first change.
Way 2: instructions and projects in every tool
ChatGPT, Claude and others let you save instructions or create projects with files. That is better, because you stop pasting every time. But it is still one copy per tool, and you keep each one up to date by hand.
Way 3: one source every tool reads
The lasting fix is to keep the brand in one place and let the tools read it from there, every time they need it. That is what MCP (Model Context Protocol) is for: ChatGPT, Claude and Cursor can connect to a source and fetch what they are allowed to read. Change the source, and every tool has the change on its next read.
What belongs in the source
- Who you are, and who you are for.
- Voice and tone, with examples.
- Visual direction: colours, type, image style.
- Rules: what must never be said or shown.
- The products and projects, with what is true right now.
In Marshi
Build it once in Marshi. Connect ChatGPT, Claude or Cursor with Brand Key, and they read the current brand every time, within what you granted. Change something in Marshi, and every connected tool has it on its next read.
