Claude GPT Launcher
Enables Codex chat to use the Claude Code harness for repository consultation and editing, routing through GPT models via a local proxy.
README
<p align="center"> <img src="Resources/AppIcon.png" width="180" alt="Claude GPT Launcher icon"> </p>
<h1 align="center">Claude GPT Launcher</h1>
<p align="center"> <strong>The Claude Code workflow. GPT models. One click.</strong> </p>
<p align="center"> <img alt="macOS 14+" src="https://img.shields.io/badge/macOS-14%2B-111827?logo=apple"> <img alt="Apple silicon" src="https://img.shields.io/badge/Apple%20silicon-native-2563EB"> <img alt="Swift" src="https://img.shields.io/badge/Swift-5.9%2B-F05138?logo=swift&logoColor=white"> <img alt="MCP" src="https://img.shields.io/badge/MCP-ready-7C3AED"> <a href="https://www.npmjs.com/package/claude-gpt-launcher"><img alt="npm" src="https://img.shields.io/npm/v/claude-gpt-launcher?logo=npm"></a> <img alt="MIT license" src="https://img.shields.io/badge/license-MIT-16A34A"> </p>
Claude GPT Launcher is a tiny native macOS app for developers who enjoy the Claude Code terminal experience and want to route it through GPT models exposed by a local Codex-compatible proxy.
Pick a repository. Pick a model. Get a real Claude Code session—without copying tokens, pasting environment variables, or leaving a proxy running forever.
[!IMPORTANT] This is an independent community project. It is not affiliated with, endorsed by, or supported by Anthropic or OpenAI. It relies on the third-party
raine/claude-code-proxyproject and may be affected by provider policy or protocol changes.
Why does this exist?
Claude Code has a distinctive interactive harness: project awareness, planning, edits, tool calls, and a terminal-first workflow. Codex-backed GPT models offer a different model family and subscription path. This project joins those two pieces behind a friendly Mac launcher and an optional MCP bridge.
Native Mac app or Codex chat
↓
Claude Code harness
↓
localhost compatibility proxy
↓
Codex / GPT models
What you get
| Surface | What it does |
|---|---|
| 🖥️ Native app | Pick any Git project and open a configured Claude Code session |
⌨️ claude-gpt |
Start the same workflow from any terminal |
| 🔌 Codex MCP | Let a Codex chat consult or edit through the Claude Code harness |
| 🧭 Model picker | Switch between configured Sol, Terra, and Luna model aliases |
| 🧹 Owned cleanup | The session stops its own proxy and removes temporary state |
Quick start
1. Install the prerequisites
Install Claude Code using Anthropic's package:
npm install --global @anthropic-ai/claude-code
Install the proxy:
brew install raine/claude-code-proxy/claude-code-proxy
Start the proxy's supported browser OAuth flow:
claude-code-proxy codex auth login
claude-code-proxy codex auth status
The login result is stored by the proxy in macOS Keychain. Never paste or commit the resulting credential.
2. Install Claude GPT Launcher
Install and build from npm:
npx claude-gpt-launcher install
Or download the Apple-notarized ZIP from the latest GitHub release, verify its checksum, and move Claude GPT.app to Applications:
shasum -a 256 -c Claude-GPT-Launcher-*.zip.sha256
3. Open a project
Open Claude GPT from ~/Applications, drag in a Git repository, choose a
model, and select Open Claude GPT.
Or skip the window:
claude-gpt "$HOME/code/my-project"
Bring the harness into Codex
The installed app bundles a local stdio MCP server:
claude_code_planperforms a read-only harness consultation.claude_code_editperforms repository edits without Bash or network tools.
Register it with Codex:
claude-gpt-launcher mcp install --enable-edits
claude-gpt-launcher mcp status --json
Then open a new Codex chat. The MCP result returns directly to that chat—no second terminal and no prompt copy/paste.
To protect sensitive repositories by remote substring:
claude-gpt-launcher mcp install \
--enable-edits \
--protected-remotes "company/production,example/private-app"
Matching repositories remain denied until a caller explicitly opts in.
Useful commands
# Check platform, dependencies, app, backend, and MCP state
claude-gpt-launcher doctor --json
# Open the app
claude-gpt-launcher open
# Remove only the MCP registration
claude-gpt-launcher mcp remove
# Remove owned app, helper, and MCP artifacts
claude-gpt-launcher uninstall
Safety model
The defaults are intentionally narrow:
- The proxy binds to a randomly selected
127.0.0.1port. - A port collision fails closed instead of reusing an unknown listener.
- OAuth material is never printed, extracted, or committed by this project.
- MCP edits require both installation-time enablement and call-time confirmation.
- The MCP excludes Bash and network tools.
- Repository roots and symlinks are canonicalized before delegation.
- Prompts, turns, output, and runtime are bounded.
- Installation and removal refuse to overwrite unrelated same-name artifacts.
[!WARNING] Claude Code is not an OS sandbox and can normally read outside its working directory. The MCP adds deny rules for common credential locations, but you should still use a disposable account, container, or VM for untrusted code.
Read the full security policy and adversarial threat model before enabling MCP edits broadly.
FAQ
<details> <summary><strong>Does this extract my subscription token?</strong></summary>
No. Authentication is initiated through the proxy's supported OAuth screen. This project never prints or asks you to paste the resulting token.
</details>
<details> <summary><strong>Is this an official Anthropic or OpenAI integration?</strong></summary>
No. It is an open-source compatibility project built on another open-source proxy. Provider policies and protocols can change, so use it with that risk in mind.
</details>
<details> <summary><strong>Why not just use Codex directly?</strong></summary>
You absolutely can. This project is specifically for people who prefer the Claude Code harness or want Codex to call that harness through MCP.
</details>
<details> <summary><strong>Does closing the session stop the proxy?</strong></summary>
Yes. Each launcher session owns a temporary proxy process and cleans it up when Claude Code exits.
</details>
Build from source
./script/install_backend.sh
./script/build_and_run.sh --install
npm test
Requirements: Apple-silicon Mac, macOS 14+, Swift 5.9+, Node.js 20+, Claude
Code, and claude-code-proxy.
Contributing
Issues, focused pull requests, documentation improvements, and security-minded reviews are welcome. Start with CONTRIBUTING.md. Please use private vulnerability reporting for security issues.
Acknowledgements
raine/claude-code-proxyfor the Anthropic-compatible local proxy.- Claude Code and Codex for inspiring a delightfully strange bridge between two excellent developer workflows.
License
MIT — see LICENSE.
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。