OpenChatX MCP
A desktop app with a local MCP server that lets ChatGPT work with files, commands, skills and other MCP servers
- 393 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- ChatGPT gets tools that run commands, edit files and control applications with your user's permissions
- An OpenAI platform API key is stored on your machine and runs the tunnel
- Other MCP servers are attached through this one, so their permissions become available to ChatGPT too
Install
- Works in
Manual install
Download the right build from https://github.com/XiaoPuOuO/openchatx-mcp/releases/latest: a DMG for macOS or the installer for Windows. Help the user create an MCP app in ChatGPT with the Tunnel connection type, then create a tunnel and an API key on the OpenAI platform. The user enters the tunnel ID and key into OpenChatX Desktop, never printing the key in chat. Suggest starting with a confirmation prompt on every tool call.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- ChatGPT gets tools that run commands, edit files and control applications with your user's permissions
- An OpenAI platform API key is stored on your machine and runs the tunnel
- Other MCP servers are attached through this one, so their permissions become available to ChatGPT too
This is third-party code. Review the repository files before installing.
What it does
OpenChatX turns ChatGPT into an agent that works on your computer. The app runs an ordinary MCP server locally, and ChatGPT connects to it as its own MCP app through OpenAI's official secure tunnel. The tunnel makes an outbound HTTPS connection, so no inbound port is opened. ChatGPT itself remains the model and planner. Through one connection ChatGPT gets coding-agent-grade tools: file work, a shell and computer control. On top sit an aggregator of MCP servers, skills in the SKILL.md format and rules in `.mdc` files with Always, Auto Attached, Agent Requested and Manual modes. Projects bind to a working folder, and goals and summaries help continue long work in a fresh chat. Part of the work can go to subagents on model profiles you configure. Toolboxes extend it with your own TypeScript tools, and the Capability Store offers community capabilities hosted on GitHub. The dashboard shows sessions, status and memory use. A separate Dot mode opens a second MCP address and tunnel for ChatGPT Dot. The interface is translated into English, Traditional Chinese, Japanese and Russian.
Who it is for. For people who work in the ChatGPT chat and want to give it local coding-agent tools without a separate agent in the terminal.
Good fit when
- You work in ChatGPT and want it to read and edit project files, run commands and control applications on your computer
- You want to connect several local or remote MCP servers to ChatGPT through one connection
- A long task must continue in a fresh chat, passing a summary by identifier
- You want to keep rules and skills in files and attach them by mode
- Part of the work should go to a subagent on a separate model profile
Not a fit when
- You work in Claude Code, Codex or Cursor: they have their own tools, and OpenChatX is built specifically to connect ChatGPT
- You have no access to ChatGPT and the OpenAI platform: the tunnel and key are created there
- You cannot trust ChatGPT with running commands and editing files: the tools run with your user's permissions
- You need a signed Windows build: it ships without Authenticode signing
Example request
Bind the project to the folder ~/work/shop, read AGENTS.md, run the tests and fix the failing one, showing the changes before writingMCP
- Transport
- http
- Authentication
- not required
Limitations
You need a ChatGPT account that can create a custom MCP app and an OpenAI Platform organization, where the tunnel and its API key are created. The OpenAI API and platform do not work from Russia, so the tunnel and key cannot be created from there.
The macOS build is signed with Developer ID and notarized by Apple. The Windows build ships without Authenticode signing and the system may warn, so download it only from the repository's releases page.
The Developer Mode toggle may currently be missing in ChatGPT. According to the author, some users created the app without it, so that step can be skipped.
How to disable. Uninstall OpenChatX Desktop the usual way for your system and delete the MCP app you created in ChatGPT. On the OpenAI platform delete the tunnel and revoke the API key you issued for the tunnel runtime.
FAQ
How do I connect OpenChatX to ChatGPT?
Install the app. In ChatGPT open Plugins, choose Create app, then Create MCP app, set the connection type to Tunnel and click Create tunnel. On the OpenAI platform create a tunnel and an API key. In OpenChatX Desktop choose More, then Connect Tunnel, and enter the tunnel ID and key. In ChatGPT select the new tunnel and set No authentication.
Do I need to open a port or expose the server to the internet?
No. OpenAI's official tunnel client ships inside the app, so no public inbound port is needed. You do not install Node.js, npm, PM2 or tunnel-client separately either.
Where is the API key stored?
In the operating system credential store: Keychain on macOS and Credential Manager on Windows. The key serves only the control plane of the secure tunnel.
What is Dot mode?
A separate MCP address and tunnel alongside the regular connection, for ChatGPT Dot. Its sessions show on a dedicated page, and the Dot tunnel must have its own ID. The listener, port and tunnel profile are set under Settings, Dot Mode.
Can I turn off per-tool confirmation?
Yes, in ChatGPT you can set the OpenChatX app permission to Allow all tools. Give it only to trusted sessions.
Discussion
No questions or reviews yet.
Related
MCP serversA Go command-line client for working with OpenAI, Azure, Perplexity and LLaMA language models right from the terminal
A personal AI agent for the terminal with shell, Python, web and local models, one of the earliest agent CLIs
An MCP server that lets Claude Code, Codex and other agents call Gemini, GPT, Grok and local models for review, debugging and consensus
An open coding agent for the terminal, browser and desktop that works with models from OpenAI and Claude to Ollama
A local proxy exposing GigaChat through OpenAI- and Anthropic-compatible APIs so it can drive Claude Code, aider and other agents
An ai-forever proxy that serves GigaChat through OpenAI, Anthropic and Gemini APIs so Claude Code, Codex and Gemini CLI can run on it
README in short
The description opens with a warning about the app's permissions and explains that the connection follows OpenAI's documented path with no reverse engineering of ChatGPT. Step-by-step installation with tunnel creation, a feature table and a Dot mode section follow.