Codex CLI Config Generator

Generate a working Codex CLI config.toml

OpenAI’s Codex CLI is configured with a TOML file at ~/.codex/config.toml plus an AGENTS.md instructions file. This tool builds both from a short form, with copy and download for each — no TOML syntax to memorize.

Install Codex CLI

npm install -g @openai/codex

Or via Homebrew: brew install --cask codex.

What the config controls

• model — the Codex model, e.g. gpt-5.2-codex or the cheaper gpt-5.1-codex-mini.
• model_reasoning_effort — minimal, low, medium, high, or xhigh (xhigh on codex-max / 5.2 only).
• approval_policy — untrusted, on-failure, on-request, or never; controls when Codex asks before acting.
• sandbox_mode — read-only, workspace-write, or danger-full-access.
• [mcp_servers.*] — Model Context Protocol servers (Filesystem, GitHub, Context7, Playwright, Postgres) configured in the same file.

AGENTS.md

Codex reads AGENTS.md from your project root (and ~/.codex/AGENTS.md globally). The optional in-browser AI writes a tailored one from a plain-English description of your project — it runs entirely in your browser via WebGPU and never uploads your description.

FAQ

Where does config.toml go?

Global config lives at ~/.codex/config.toml; a project can override it at <project>/.codex/config.toml.

Does it send my data anywhere?

No — the config is generated locally in your browser, and the optional AGENTS.md writer also runs locally.