ChatGPT Local Terminal Plugin

ChatGPT chat has a separate, near-unlimited quota from Codex and Work—but it can’t control your computer like Codex. Until now.

An unofficial, self-hosted MCP server that lets ChatGPT run commands on your computer through an OpenAI Secure MCP Tunnel.

It works similarly to Codex: ChatGPT can run commands, read output, manage long-running processes, send input, and work inside a chosen project directory.

Codex and ChatGPT Work have limited shared quotas, while the regular Chat interface pulls from a seperate quota pool. This lets you use high end models like Sol without cutting into your limited Work and Codex usage limits. Usage limits and model availability depend on your OpenAI account and may change; this tool does not guarantee extra or unlimited quota.

How it works

ChatGPT
  → OpenAI Secure MCP Tunnel
  → tunnel-client
  → local MCP server
  → commands on your computer

The folder where you run start.sh becomes the workspace.

Installation

Currently supported on Mac, untested on Linux. Winodws support available via WSL (manual installation, untested).

1. Clone the repository

git clone https://github.com/Angad-D/ChatGPT-Local-Terminal-Plugin-Unofficial.git
cd ChatGPT-Local-Terminal-Plugin-Unofficial

2. Run the installer

./install.sh

The installer creates a local Python environment, installs the MCP server, and downloads the latest official OpenAI tunnel-client.

3. Create a Secure MCP Tunnel

Create your own tunnel and runtime API key in OpenAI Platform.

Tunnel settings:

https://platform.openai.com/settings/organization/tunnels

Runtime API keys:

https://platform.openai.com/settings/organization/api-keys

4. Configure .env

Edit:

nano .env

Add:

CONTROL_PLANE_TUNNEL_ID=tunnel_your_id
CONTROL_PLANE_API_KEY=sk-your-runtime-api-key
MCP_SERVER_URL=http://127.0.0.1:8000/mcp

Do not commit .env.

5. Choose an access mode

Edit:

nano config.toml

Recommended:

sandbox_mode = "workspace-write"
network_access = true
fail_if_sandbox_unavailable = true

Available modes:

sandbox_mode = "read-only"
sandbox_mode = "workspace-write"
sandbox_mode = "full-access"

full-access disables filesystem sandboxing and should only be used when necessary.

Linux users need Bubblewrap for sandboxed modes:

sudo apt install bubblewrap

Connect to ChatGPT

Start the tool from the project directory you want ChatGPT to access:

cd ~/Documents/projects/my-project
/path/to/ChatGPT-Local-Terminal-Plugin-Unofficial/start.sh

Then in ChatGPT:

1. Enable developer mode.
2. Create a custom app/plugin.
3. Choose Tunnel as the connection type.
4. Paste in the tunnel id.
5. Choose "No Auth"
6. Save and enable.
7. Change the plugin settings to "Always Ask"

If it doesn't let you type in the tunnel id, and asks you to create a new one after you have already made one (previous step), click "Create" in the bottom right corner of the popup (for creating the plugin), and the create option for the tunnel will switch to a text box in which you can type in your tunnel id.

The exact ChatGPT interface may change over time.

Usage

Example prompts:

List the files in this workspace.

Run the project's tests and fix any failures.

Start the development server and keep it running.

Read the latest output from the development server.

Stop the running development server.

Commands run in separate processes rather than one permanently shared shell. ChatGPT should specify the correct working directory for each command.

Stopping the tool

Normally, press:

Ctrl+C

If processes remain:

/path/to/chatgpt-terminal-mcp/stop.sh

Security warning

This tool gives ChatGPT the ability to execute commands on your computer.

Commands may:

• modify or delete files;
• install or execute software;
• make network requests;
• expose sensitive information through command output;
• consume system resources.

Recommended precautions:

• Use workspace-write.
• Run it only from a project directory you are comfortable exposing.
• Do not run it from your home directory or filesystem root.
• Keep important projects under version control.
• Do not send passwords, API keys, recovery codes, or MFA codes through ChatGPT.
• Read all commands before allowing ChatGPT to run them (set permissions to "Always Ask")
• Stop the server when it is not being used.

No sandbox is a complete security guarantee. full-access gives commands the same filesystem access as your user account.

Unofficial project

This is an independent, unofficial open-source project.

It is not affiliated with, endorsed by, approved by, or maintained by OpenAI.

Each user must create and secure their own:

• Secure MCP Tunnel;
• runtime API key;
• custom ChatGPT developer app/plugin.

ChatGPT plans, models, usage limits, developer mode, and MCP availability vary and may change.