Best way to migrate my long-term ChatGPT context and project knowledge to Claude?
Hey, I’m planning to move most of my ongoing development work from ChatGPT/Codex over to Claude, especially Claude Code/Claude Design for UI since ui stuff is just not good on GPTs end.
I’m not just talking about moving a single repository. Over time, ChatGPT has accumulated a pretty large amount of context about me, how I work, my preferences, my projects, infrastructure, naming conventions, design direction and the way I want agents to approach tasks.
My projects include multiple websites and web design projects, internal dashboards, Next.js/React apps, automation workflows, self-hosted services, APIs, a mail server, server infrastructure, deployment setups and various experimental tools. Some of these projects are interconnected, so having the broader context available to the agent is actually pretty useful.
I already have project-specific rules, skills and instructions set up, and I’m planning to do the same properly on the Claude side rather than just throwing one gigantic prompt at it.
What I’m mainly trying to figure out is the best way to transfer the more persistent “knowledge about me” that ChatGPT has built up over time.
Ideally I’d like Claude to understand things like my general development preferences, how I structure projects, my infrastructure, recurring services and domains, design preferences, commonly used stacks, deployment conventions, the different brands/projects I work on, which projects belong together, how I prefer agents to work, and important decisions or problems we already solved in the past.
I don’t necessarily want to export years of chats and dump everything into Claude because that seems like a good way to create a huge amount of noisy and outdated context.
I’m thinking it would make more sense to create some kind of structured knowledge base with a global profile/context file, infrastructure documentation, project-specific context files, decision logs and then separate rules/skills that Claude Code can load when relevant.
Has anyone done a migration like this?
How would you structure it so Claude gets as much useful context as possible without constantly loading thousands of irrelevant tokens?
Also interested in whether you would use CLAUDE.md files, separate Markdown knowledge files, MCP, Claude Projects, skills, or some combination of those.
Basically I want to turn years of accumulated ChatGPT context into a clean, maintainable “second brain” for Claude rather than starting from zero on every project.