Settings and MCP Can "Move House" Too — Codex 0.145's /import Brings Your Entire Cursor and Claude Code Setup Along

When you switch coding agents, the thing that drags on you until the very end is all the configuration you've built up over time. The MCP servers you connected, the plugins you installed, the commands you reach for again and again, the notes you've taught it project by project — rebuilding all of this from scratch in a new tool is a far heavier lift than you'd expect. Codex 0.145.0, released by OpenAI on July 21, has consolidated this "packing for the move" into a single command.

An entry point that lets you move in with all your belongings

The headline is that the existing /import command now covers far more. On top of what it could already pull in, it can now import your Cursor and Claude Code settings. And it isn't just your preference values that come along — the assets that form the foundation of the agent are all in scope.

• MCP servers: connection definitions for external tools and data
• Plugins: your full set of extensions
• Sessions: past conversations
• Commands: shortcuts for invoking routine work
• Per-project memory: the assumptions and conventions you've taught it for each repository

In short, the "personal workspace" you've crafted in Cursor or Claude Code can be brought into Codex almost as-is. The initial setup effort that used to stand in the way of switching is cut down dramatically.

Toward choosing by fit and cost, not by feature gaps

This move matters because it's about more than mere convenience. The easier it becomes to port your configuration, the thinner the switching cost (lock-in) that once tied you to a tool. Developers move closer to a decision of "I'll pick this because it suits the task at hand" rather than "I'll keep using this because I can't abandon my current setup."

In fact, the Cursor community had voiced requests to carry sessions and history back and forth between Claude Code, Codex, and Cursor. Now that vendors are converging on the same "agentic coding" design, the axis of differentiation is shifting from raw features to price, model fit, and how well a tool meshes with your existing assets. The expanded import capability can be read as a sign that Codex has entered that competition.

Multi-agent goes from "experiment" to "everyday use"

Another milestone is the stabilization of Multi-Agent V2, previously an opt-in feature. This update brings the machinery for running multiple subagents into a form that can withstand real-world use. Specifically, you can now choose the model, reasoning depth, and concurrency for each subagent, and assigning roles and moving between agents has become easier to handle.

Splitting one large task into a researcher, a writer, and a checker and running them in parallel — the groundwork is now in place to keep that kind of division of labor going day to day while controlling it through configuration. Cost design such as letting a cheaper model do the legwork and handing only the critical parts to a higher-tier model also becomes realistic now that you can specify models individually.

Changes that landed alongside

• Paginated history (experimental): supports resuming, searching, and naming past threads, plus subagents and memory, making history easier to navigate even over long-term use.
• Amazon Bedrock support (experimental): adds Bedrock login and custom endpoints/authentication, with GPT-5.6 Sol as the default model — broadening the option of running Codex on your own cloud.
• Voice and realtime conversation: adds voice input and tool output (supporting common local audio formats), plus streaming realtime conversation V3.
• Terminal improvements: incremental Markdown rendering reduces display lag, and safely openable visualization links are shown in the terminal.

What to check behind "bring it all along"

The ease of migration is progress worth welcoming, but it pays to be careful that what you're importing includes MCP servers and plugins. These are mechanisms that carry connections to the outside and execution privileges, and if you bring them in unvetted on the grounds that "they worked in my old environment," you'll inherit any oversights in permissions or the supply chain right along with them. In fact, tactics that try to manipulate an agent through fake error reports or external content have been reported repeatedly over the past six months. Migration is precisely the right moment to take stock of which servers you've granted what. Treating it that way is the safe approach.

Also, the fact that switching gets easier means, conversely, that competition and turnover among tools grows fiercer. Behind the convenience, the decision of which environment your team standardizes on arguably becomes more necessary than ever. Start by staying mindful of how you handle experimental features, try migrating on a small project, and verify the permissions of each imported setting one by one — that bit of extra effort turns lightness into safe lightness.

References: Codex 0.145.0 release notes (GitHub) / Codex changelog (official) / Codex Updates - Releasebot / Native Agent History Import (Cursor Community Forum)