Skip to content
Summer sale. 50% off Pro, $19.50 for life. Desktop app: Apple Silicon Mac only.SUMMER50
CodexUse

Codex settings in config.toml: models, reasoning, sandbox, MCP

SettingsUpdated August 2026 · 5 min read

Codex keeps its settings in config.toml, and you edit that file yourself. CodexUse does not put a form in front of it: there is no config panel and no raw editor in the app. What CodexUse does give you is separation — every profile runs against its own Codex home, so every profile carries its own config.toml and one never overwrites another.

Manual today: every setting below is a line you type into a text file.

What lives in config.toml

  • Model and review model: which model Codex runs and which one reviews.
  • Reasoning effort and summary style: model_reasoning_effort and model_reasoning_summary.
  • Verbosity: model_verbosity.
  • Approval policy: approval_policy, for example never or on-request.
  • Sandbox mode: sandbox_mode, for example read-only, workspace-write, or danger-full-access.
  • MCP servers: one [mcp_servers.NAME] block per server.

Codex’s own documentation is the reference for the full key list and every accepted value.

model = "gpt-5.1-codex"
approval_policy = "on-request"
sandbox_mode = "workspace-write"
Careful with that pair: approval_policy = "never" together with sandbox_mode = "danger-full-access" removes both the approval prompt and the sandbox.

One config.toml per profile

This is the part CodexUse actually changes. Each profile points Codex at its own home directory, so:

  • A loose sandbox setting on a throwaway profile does not follow you into a client profile.
  • MCP servers stay with the profile you added them to. See the MCP server setup guide.
  • Switching profiles switches the whole configuration with it, with no file to swap by hand.

Custom model providers are the exception

Codex can point at a non-OpenAI, OpenAI-compatible endpoint through its own configuration, but by hand that means editing config files and keeping a credential near them. A custom-provider profile does that setup for you instead, with the credential in the macOS Keychain and an isolated Codex home per profile.

Editing without breaking anything

  • Codex reads the file when a session starts. Save, then start a new session to pick the change up.
  • Keep the file valid TOML. Change one key at a time and confirm the next session still starts.
  • Cloud Sync pushes and pulls config.toml along with your profiles, and a pull replaces the local copy. See the cloud sync guide.

Does CodexUse have a settings form for this?

No. There is no config panel and no raw config.toml editor in the app. You edit the file yourself. CodexUse manages accounts, profiles, Codex App windows, and quota around it.

Do I need Pro to change config.toml?

No. config.toml is a Codex file on your own disk. A CodexUse license covers CodexUse features such as auto-roll, Accounts Pool, Telegram alerts, and Cloud Sync.

Which Codex CLI does this use?

CodexUse drives the Codex CLI already installed on your machine, so install @openai/codex first if codex is not already on your PATH.

Which platforms are supported?

Current public desktop releases target Apple Silicon macOS. For Linux or SSH workflows, use the optional CodexUse CLI. The desktop product is the native app.