Codex v0.111.0: Fast mode by default, structured MCP, and a Windows app lands
OpenAI Codex v0.111.0 turns on Fast mode by default, tightens MCP integration, expands image workflows, and debuts an early Windows app with mixed first impressions. The release makes Fast mode the default, adds dynamic local .js/.mjs imports in js_repl, and shares enabled MCP apps and skills with the model at session start ([release notes](https://github.com/openai/codex/releases/tag/rust-v0.111.0)). App‑server v2 now exposes MCP elicitation via a structured request/response API, image workflow handling is expanded, and resuming threads preserves git context and app enablement. A Windows build of the Codex app is out, with community reports of stutter and some confusion about how to quit the app ([Windows announcement](https://community.openai.com/t/the-codex-app-is-now-on-windows/1375704), [stutter report](https://community.openai.com/t/codex-app-is-kinda-stuttery-on-windows/1375836), [quitting thread](https://community.openai.com/t/how-does-one-quit-the-codex-application-on-windows/1375813)). There is also feedback asking for message editing and proper undo in the VS Code plugin ([plugin request](https://community.openai.com/t/codex-vs-code-plugin-add-message-editing-and-proper-undo/1372929)). Plan for app‑server v2’s structured MCP flow in clients and validate Fast mode’s performance, cost, and rate limits in CI or agent workflows.