Codex CLI Routes Through ChatGPT Web, Bypassing API Costs at 8,500 Stars
A new open-source launcher routes Codex through your ChatGPT Web session, letting Pro subscribers use their chat quota for agentic coding instead of API credits.
- codex-chatgpt-web exposes ChatGPT Web models, including Pro, as native Codex models via browser automation.
- Uses your ChatGPT subscription quota instead of Codex or API credits, hitting 8.5k GitHub stars.
- Full harness mode wires ChatGPT tool calls back to Codex files and terminal through MCP.
- Three modes: Browser-only, Full harness with automation, and manual Zero Risk paste-and-send.
- Context windows up to 270,000 tokens with experimental 3x mode on Plus accounts.
- Unofficial and fragile: relies on UI selectors, raises real security and terms-of-service concerns.
Codex ChatGPT Web routes Codex CLI through a ChatGPT session
The TypeScript project codex-chatgpt-web has crossed 8,500 GitHub stars. It adds a provider to OpenAI’s Codex CLI, the terminal coding agent that can inspect repositories and run approved commands, then routes model requests through a signed-in ChatGPT Web session.
The project says requests count against the user’s ChatGPT Web allowance instead of Codex quota or API credits. Developers can select models available through their ChatGPT plan, including Pro where enabled, while retaining Codex features such as streaming, task context, image support, and local tools. The trade-offs include browser automation, access to sensitive session data, UI-dependent reliability, and uncertain policy status.
Inside the browser bridge
The launcher includes its own browser and runtime. After the user signs in to ChatGPT, it receives requests from Codex, submits them through the web interface, and streams responses back to the terminal. Conversations remain associated with the current Codex task as its context grows.
Full harness mode connects ChatGPT to Codex through the Model Context Protocol, or MCP. MCP gives the model a standard way to request actions such as reading files, editing code, and running terminal commands. Codex executes those actions inside its local sandbox and applies its usual approval flow.
Tool calls travel through OpenAI’s official tunnel client. The connection starts from the local machine, so it requires no inbound port, public IP address, or router forwarding. Full harness setup requires three steps:
- Enable Developer Mode in ChatGPT.
- Create a connector named
Codex Native2. - Start the launcher and attach the connector to the current Codex task.
Automation comes in three modes
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
