Codex 26.608: Import Your Claude Code Config with One-Click Migration

A new migration flow targets Claude Code users directly. Codex app 26.608 introduces a guided "Migrate to Codex" onboarding experience that imports your existing configuration from Claude Code and Claude Cowork. This is distinct from the CLI desktop-handoff feature covered in yesterday's June 8 release — that update bridged the terminal and desktop surfaces; this one is specifically about lowering the cost of switching coding agents entirely.

Six config types transfer in the migration. The flow imports: system prompts, custom skills and shortcut commands, chat history from the past 30 days, MCP server configurations, hooks, and sub-agent settings. That covers the majority of the operational setup most teams accumulate in a coding agent over weeks of use — the kinds of configurations that historically made switching tools feel expensive even when a better option existed.

Permissions require a manual review after migrating. OpenAI explicitly recommends reviewing permissions after the import — the permission systems between Codex and Claude Code are not interchangeable, and configurations that carried over may not map cleanly to Codex's permission model. Operators should treat the migration as a starting point, not a finished port: audit what transferred, verify scope, and don't assume identical behavior. This caveat applies especially to hooks and sub-agent settings, where permission boundaries differ most between the two platforms. For a full breakdown of how Claude Code's permission model works before migrating, see the Cursor vs Claude Code comparison.

The plugins screen was also revamped. The same 26.608 release brings separate marketplace tabs, category filters, and keyboard navigation to the plugins screen. These are UX improvements rather than capability additions, but they matter for teams evaluating Codex's plugin ecosystem — easier discovery lowers the time to find and configure relevant plugins. Settings search was also expanded in this release.

CLI 0.139.0 adds web search inside code-execution mode. The companion CLI release (also June 9) ships standalone web search in code mode — the first native web-search capability in Codex's code-execution context. This includes web search from nested JavaScript tool calls, meaning agents running code can now reach the web mid-execution without leaving the code-mode environment. Previously, web search required switching contexts. For teams building on the best AI coding tools for agentic workflows, this closes a gap that distinguished Codex from tools with more integrated search capabilities.

The competitive signal here is deliberate. Building a migration importer that specifically names Claude Code and Claude Cowork is not an incidental feature — it's OpenAI identifying Claude Code as the primary reference point for developers it wants to convert. The specificity of what transfers (MCP configs, hooks, sub-agents) signals that OpenAI understands how Claude Code users actually set up their environments. For teams currently evaluating coding agents, this materially lowers the cost of running a side-by-side trial: you don't need to rebuild your config from scratch to give Codex a fair test.