From the team behind aidrop.it — one workspace to build, host, and keep changing your code.
You shouldn't have to introduce yourself to your own tools. Yet every time you open a new AI session — Claude Code for a refactor, Codex for a review, ChatGPT for a proposal — you start at zero: same project background, same preferences, same constraints, retyped again. The fix is to stop storing context inside tools and keep it in one portable memory that every tool connects to over MCP.
The re-explaining tax is real. If you spend five minutes per session restoring context across three tools, that's hours per month — and the sessions where you skip it are the ones where the AI quietly gets things wrong.
Why your context doesn't transfer
Every AI tool keeps memory in its own silo. ChatGPT has its built-in Memory, Claude Code reads CLAUDE.md, Codex reads AGENTS.md, and none of them can see the others. The knowledge you accumulate in one tool — preferences, decisions, project facts — is locked to that tool, and often to a single machine.
Silos also mean divergence. You update a decision in one place and the other tools keep acting on the old version. The more tools you use, the more copies of "what's true" you maintain by hand.
What portable memory looks like
Portable memory is a single store that lives outside any one tool and speaks MCP — the open protocol that Claude Code, Codex, ChatGPT, and most modern agents already support. Each tool connects as a client: it can search the memory, read pages, and write new facts as you work. Add a tool, and it knows everything the others learned. (For how this layer relates to context windows and RAG, see AI agent memory explained.)
How it works in practice
In practice, a shared memory is connected once in each tool as an MCP server. As you work, agents record what they learn; a merge step folds it into a versioned store, so the memory stays curated instead of becoming a log dump. When any tool needs context, it searches the memory and pulls only the relevant pieces.
The effect day to day: you tell Claude Code about an architecture decision on Monday, and ChatGPT already knows it when drafting the client update on Tuesday. Versioning means you can also see how a decision evolved — not just its latest state.
Switching tools becomes free
Once context lives outside the tool, tools become interchangeable. You pick whichever model is best for the task — or cheapest, or has limits left — without paying the re-onboarding cost. Your leverage stops being "the tool I've trained" and becomes "the memory I've built", which compounds across every tool you'll adopt next. The same logic applies to personal agents — see how to make OpenClaw smarter with a memory.
FAQ
What is MCP? Model Context Protocol — an open standard that lets AI tools connect to external services like databases, APIs, and memory stores. Claude Code, Codex, ChatGPT, and many other clients support it natively.
Can't I just paste the same notes into every tool? You can, but you become the sync mechanism. Every update has to be re-pasted everywhere, and stale copies cause silent mistakes. A shared memory updates once, everywhere.
Does shared memory mean my data trains someone's model? Not by itself — a memory store you run or control stays out of model training, but check each provider's terms. Tools read and write it on your behalf; it's not training data.
One workspace to build, host, and keep changing your code
The repository, what the project knows, the rules a change has to follow, and the path to an address — kept together, and reachable by the coding agent your team already uses.