ChatGPT To Codex
A local MCP runtime for macOS and Windows that enables ChatGPT to read, edit, search, run tests, and capture screenshots in your local project folder while keeping your source code on your machine.
README
<p align="center"> <img src="assets/readme-hero.png" alt="ChatGPT To Codex local coding runtime" width="100%" /> </p>
ChatGPT To Codex
Give ChatGPT real local coding hands.
ChatGPT To Codex is a local MCP and Actions runtime for macOS and Windows that lets ChatGPT work inside the project folder you choose: read files, search code, apply patches, run tests, launch E2E checks, and send back screenshot proof.
Your source stays on your machine. ChatGPT connects to the local app you run. You choose the workspace, approve the token, and keep control of what gets edited.
Download v0.1.2 · Beginner installation guide
Help us get this in front of more builders: star the repo if you want ChatGPT to stop talking about code and start safely doing the repo loop.
Why It Exists
ChatGPT is great at reasoning, but web chat alone cannot reliably inspect your local repo, run your local tests, or prove what the UI actually looked like. ChatGPT To Codex fills that gap:
- local project selection instead of uploading a source tree
- guarded file reads and hash-checked patching
- allowlisted local commands for tests and checks
- macOS/Windows app, window, and browser screenshot capture for visual E2E proof
- temporary or fixed HTTPS connector URL for ChatGPT web
- OAuth-style owner-token approval so random clients cannot just attach
- multilingual menu bar app for non-English users
The mental model is simple:
ChatGPT thinks. Your computer acts. You review the result.
Current Release
| Platform | Status | Package |
|---|---|---|
| macOS | Public release | chatgpt2codex-0.1.2.pkg |
| Windows | Public release | chatgpt2codex-0.1.2-windows-setup.exe |
| Linux | Developer path only | Not published |
Why PKG Instead Of DMG?
For this release, PKG is the better fit. A DMG is nicer for drag-and-drop apps, but this app needs to install a menu bar runtime under Applications, bundle Node/cloudflared helpers, and run a non-blocking post-install Doctor. PKG gives beginners a clearer "install and open" path. A signed/notarized DMG can still be added later for a more consumer-style download.
Current macOS package SHA-256:
d9226ba10d9a24a121a0ec6c7a6c79f81d55b3bb60ddc5acd13240c2d17f7821 chatgpt2codex-0.1.2.pkg
What ChatGPT Can Do With It
Once connected, ChatGPT can operate like a practical coding agent over a trusted project:
- list local projects and select the active one
- read repo rules before editing
- search code and read exact line slices
- create files and apply patches
- run project commands and tests
- start a dev server and wait for a URL
- open a browser URL or installed desktop app
- capture macOS/Windows E2E screenshots
- return inline screenshot previews through Actions
- save generated image assets into the repo
- summarize diffs, blockers, and verification evidence
The standout workflow is:
Run the E2E test, open the app, capture screenshots, and show me proof.
For web apps, ChatGPT To Codex can capture browser regions. For desktop apps
such as Tauri apps, it can open the built app window and capture top/middle/bottom
views. The one-shot e2e_test_and_show_screenshot action returns inline
imageMarkdown results so you can inspect the screen without digging through
local folders.
Install In 5 Minutes
Full beginner guide: docs/INSTALL.md
macOS short version:
- Download
chatgpt2codex-0.1.2.pkgfrom the latest release. - Open the installer.
- If macOS blocks the unsigned package, Control-click it, choose Open, and confirm in System Settings -> Privacy & Security if needed.
- Open ChatGPT To Codex from Applications.
- Open Settings... from the menu bar icon.
- Choose a project folder.
- Enable ChatGPT web connector if you want ChatGPT in the browser to connect.
- Click Start MCP.
- Click Copy Connector URL.
- Register that
/mcpURL in ChatGPT Apps / Connectors and approve with the Owner Token shown by the app.
Windows short version:
- Download
chatgpt2codex-0.1.2-windows-setup.exefrom the latest release. - Double-click the installer.
- If Windows SmartScreen warns, choose More info -> Run anyway only if the file came from this GitHub release.
- Launch ChatGPT To Codex.
- Open the tray icon settings, choose your project folder, enable the ChatGPT web connector if needed, then click Start MCP.
- Copy the
/mcpConnector URL and approve it in ChatGPT with the Owner Token.
Keep the Owner Token private. Treat it like a password.
First Prompt To Try
Use ChatGPT To Codex. Select my project, read the README and package scripts,
run the safest available check, then summarize the result with exact evidence.
Then try a visual proof flow:
Use ChatGPT To Codex to run the app E2E, capture screenshots, and show the
passing screenshot set inline before you say it is done.
Safety Model
ChatGPT To Codex is designed for trusted local development, not arbitrary public automation.
- It runs locally on your computer.
- It defaults to loopback-only networking.
- ChatGPT web requires an explicit connector/tunnel mode.
- File operations are scoped to the selected project.
- Patch application uses line/hash context.
- Owner Token approval is required for remote Actions access.
- Secret-looking values are redacted from tool output.
- Destructive, network, and sensitive operations remain approval-gated.
Do not expose the connector URL publicly unless you understand the tunnel and token model. Do not paste Owner Tokens into issues, screenshots, or shared logs.
Supported Languages
The desktop app can follow the system language and currently includes UI strings for English, Korean, Japanese, Simplified Chinese, Traditional Chinese, Spanish, French, German, Brazilian Portuguese, Italian, Dutch, Polish, Russian, Turkish, Vietnamese, Indonesian, Thai, Arabic, Hindi, and Ukrainian.
The install guide currently includes Korean, English, Japanese, and Simplified Chinese. More documentation languages are welcome.
Windows Status
Windows now has a public beginner installer. It includes the tray launcher, owner-token setup flow, ChatGPT web connector settings, stale runtime cleanup, and Windows E2E screenshot proof. See docs/INSTALL.md and windows/README.md for the full Windows guide.
Repository Contents
This public repository is intended to contain only the product source, public documentation, assets, scripts, and published installer artifacts. Local agent state, personal automation rules, generated memory, hooks, private MCP config, build output, and machine-local logs are ignored.
If you see local-only files in a clone, they came from your machine, not from the public repo.
Build From Source
For developers:
npm ci
npm run typecheck
npm test
npm run build
Build the macOS package:
npm run macos:package
The packaging script creates a .pkg under build/macos/. Published packages
are copied to installers/macos/.
Star Pitch
If this saves you one "copy this patch, paste it in terminal, now run tests, now send me a screenshot" loop, give it a star. The goal is simple: make ChatGPT useful for real local development without turning your project into a cloud upload.
Built by ezBuilder.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。