Connect Claude Code, Codex and ChatGPT to one understanding
22 October 2026
Naetive is built so that the agent does not matter. Claude, Claude Code, ChatGPT, Codex. Whichever one you open, it should inherit the same understanding of your project. Here is how to connect each one, and what happens when you have no connection at all.
One understanding, many agents
The understanding of your project lives in one place: Naetive, plus the .naetive folder you can keep in your repo. Every agent connects to that same understanding rather than building a private one. So the picture Claude Code has of the project is the picture Codex has, is the picture ChatGPT has.
You are not syncing between tools. There is nothing to sync. They all read from and write to the same thing.
Connecting over MCP
Naetive speaks the Model Context Protocol, the shared way agents connect to outside context and tools. Each agent connects to Naetive as an MCP server, once, with its own key from the Connect page. After that, it can read the project's understanding and commit new understanding back as part of its normal work. What differs between agents is only how you add the server.
Claude Code. Run the install line from the Connect page in a terminal, from any folder. On macOS, Linux and WSL that is curl -fsSL https://naetive.ai/install | bash, and Windows has a PowerShell line. Your browser asks you to approve the computer once. Then open Claude Code in your project folder and run /mcp to check naetive is listed. Claude Code works on your real files directly and keeps the .naetive folder in your repo current. Full steps: Connecting Claude Code.
Codex CLI. Add a naetive server to ~/.codex/config.toml, using the block the Connect page shows with Codex's own key in it. Restart Codex and run /mcp. Full steps: Connecting Codex CLI.
ChatGPT. On chatgpt.com, turn on Developer mode in settings, then create a custom MCP server with the URL from the Connect page and no authentication, since the key is inside the URL. It needs a paid ChatGPT plan. Full steps: Connecting ChatGPT.
Claude on the web and Claude Desktop. Add a custom connector in settings and paste the URL from the Connect page. It needs a paid Claude plan. Full steps: Claude on the web and Claude Desktop.
A chat agent such as ChatGPT or Claude on the web cannot see files on your computer by itself. To let it work on them, share the project and start the tunnel, a small program the install line puts on your machine. How to connect compares the paths side by side.
Switching from Claude Code to Codex tomorrow is the same step again, to the same understanding.
For a clean clone with no connection
You do not always have a live connection. A fresh checkout on a new machine, an agent that has never heard of Naetive. That is what the .naetive folder is for.
Because the understanding is committed into the repo, an agent with no MCP connection at all can still read it. Run naetive export in the repo and it adds a short pointer to your AGENTS.md, and to your CLAUDE.md if you have one, so agents read the folder on startup. If you download the folder from the Understanding page instead, it includes a HOW-TO-USE file with the lines to add yourself. Clone the code, inherit the understanding, no account required. More in The .naetive folder.
What you get
Point any agent at the project and it already knows what the project is for, how it is built, which decisions are settled and which are open. Switch tools mid-project and nothing resets. Add a second agent and it inherits what the first one learned.
One understanding. Every agent.