Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/XxYouDeaDPunKxX/chatgpt-local-agent-mcp/llms.txt

Use this file to discover all available pages before exploring further.

ChatGPT Local Agent MCP is a local MCP server that bridges ChatGPT to your Windows machine. Instead of copying files, errors, and terminal output into the chat window, you let ChatGPT work directly with the same local reality you are — reading files, running commands, inspecting Git state, opening browser sessions, and operating the desktop — all under your supervision.

Introduction

Understand what this project does and when to use it

Quickstart

Install the server and get your first health check passing

Path A — Local Smoke Test

Prove the server starts before connecting anything remotely

Path B — Remote Connector

Connect ChatGPT through a secure authenticated HTTPS endpoint

Tool Reference

Browse every MCP tool across filesystem, shell, browser, and desktop

Security & Auth

Understand the two OAuth layers and how scopes protect tool access

​
How it works

1

Install and build

Run the PowerShell installer from the extracted source folder. It copies the app into %LOCALAPPDATA%\chatgpt-local-agent-mcp, installs Node dependencies, and builds the TypeScript source.
2

Run the local smoke test (Path A)

Start the server bound to 127.0.0.1:8789 with AUTH_REQUIRED=false. Open the dashboard at http://127.0.0.1:8789/dashboard and confirm the health endpoint responds.
3

Configure OAuth and a Cloudflare Tunnel (Path B)

Set up a GitHub OAuth App for identity, supply ChatGPT connector OAuth values, and route your public HTTPS hostname through Cloudflare Tunnel to the local server.
4

Connect ChatGPT

Add the MCP endpoint https://your-public-host.example/mcp as a connector in ChatGPT. Authorize via GitHub login, and ChatGPT gains supervised access to your local tools.
Always complete Path A before attempting Path B. The remote connector adds OAuth, tunneling, and public exposure — there is no reason to debug those layers while the local server itself is still untested.

​
What ChatGPT can do

With the server running and the right scopes granted, ChatGPT can help you:
  • Inspect directory trees, read files, apply patches, and write changes back to disk
  • Run build commands, tail logs, and wait for services to become available on a port
  • Check Git status, review diffs, and create commits
  • Open Playwright browser sessions, navigate pages, capture screenshots, and inspect console/network output
  • Attach to an existing browser via CDP
  • List open windows, take desktop screenshots, perform mouse clicks, and type keyboard input
Start with narrow scopes (mcp:read only, or add mcp:git) and expand from there. The scope system lets you give ChatGPT exactly the access it needs for a task without opening the entire tool surface.

Build docs developers (and LLMs) love

Get started for freeTalk to us