how to turn ChatGPT into Codex through an MCP connector

devspace is an MCP connector that “turns ChatGPT into Codex.” The tool was built on a Codex hackathon in one day and solves a specific practical problem: GPT-5.5 Pro is the best model for planning and researching code, but it is not available directly in Codex.

Repository: github.com/Waishnav/devspace

Description: "Turn ChatGPT into Codex" - 27 stars, 3 forks at the time of writing.

## What problem does devspace solve

To understand the value of a tool, you need to understand the architectural feature on which it is built.

OpenAI offers two separate products with different models and different rate limits

ChatGPT is an interface with access to GPT-5.5 Pro. This is a powerful universal model, perfectly suited for planning, codebase research, architecture analysis and task setting. But ChatGPT itself doesn’t have access to your local file system.

**Codex is a code writing agent. Works with your local environment, reads files, runs commands, makes commits. But the main recommended model in Codex is GPT-5.5, not GPT-5.5 Pro. For ultra-fast tasks, ChatGPT Pro subscribers have access to the GPT-5.3-Codex-Spark model in the research preview.

A key finding that underlies devspace: ChatGPT and Codex have separate rate limits. This means that you can use both products in parallel – GPT-5.5 Pro for planning and research in ChatGPT, and Codex for direct code writing – and in doing so, effectively double the amount of work available without hitting a common limit.

devspace implements this pattern technically: through the MCP server, it opens the local dev environment for ChatGPT, after which ChatGPT with GPT-5.5 Pro gets the ability to see project files and transfer tasks to Codex.

## How it works technically

The devspace architecture consists of several parts:

MCP server - receives requests from ChatGPT and provides tools to work with the local developer environment. The server runs locally and acts as a bridge between the ChatGPT (cloud) and the file system (locally).

Exposure to the dev environment - after running the ChatGPT server, it accesses the context of your project: files, directory structure, git history. This is something that is usually only available to Codex CLI when working directly in the terminal.

Transfer of tasks to Codex – After GPT-5.5 Pro in ChatGPT has done planning and research, the task is sent to Codex for execution. Codex receives a predefined task and context, rather than working from scratch.

Structure of repository:

devspace
─ .github/workflows/ — CI/CD
Docs/ - Documentation
Scripts/ - Auxiliary scripts
а ─ src/ - server source code
ных─ .env.example – template of environment variables
. ─ README. md

## Workflow: What it looks like in practice

The workflow with devspace is divided into two distinct stages according to the principle of “the best model for each task”.

Stage 1 – ChatGPT Planning (GPT-5.5 Pro)

ChatGPT with a connected devspace MCP server sees your project. Here you are asking high-level questions:

• “How is the authorization in this project?”
• “Where better to add a new caching module?”
• What files should I change to add support for WebSocket

GPT-5.5 Pro explores the codebase, forms a change plan, determines the list of files for editing.

Stage 2 - Execution in Codex

The completed plan and task list are transmitted to Codex. Codex no longer spends tokens on research – it gets a clear task and starts writing code, running tests and commits.

The practical result: You get a GPT-5.5 Pro for planning and a handover in Codex for execution, using separate rate limits as a way to double the effective bandwidth.

## Why GPT-5.5 Pro Is Important for Planning

According to devspace, “GPT 5.5 Pro is the best model for planning and research.” Why is it not available in Codex? devspace is the answer to this question in the form of a working tool.

The difference between GPT-5.5 and GPT-5.5 Pro is manifested precisely in tasks where depth of reasoning is needed: understanding the architecture of an unfamiliar project, analyzing complex dependencies, formulating an exact technical task. These are the tasks that precede code writing – and this is where the benefits of a more powerful model are maximized.

## Context: Separate Rate Limits as an Architectural Feature

According to official OpenAI documentation, ChatGPT and Codex work on the same account, but some settings and connected services can be transferred between them – for example, if you connect Google Drive to ChatGPT, it will be available in Codex. The limits of use of these two products are separate.

It’s not a bug or a loophole — it’s an architectural feature of how OpenAI separated the two products. Devspace uses it consciously: instead of spending the entire Codex limit on research and planning, research is done in ChatGPT (its limit), and Codex spends its limit solely on writing code.

Use of Codex, ChatGPT for Excel and Workspace Agents is counted in one agency limit. The number of messages that can be sent within these limits varies depending on the size and complexity of the tasks – small scripts can consume only a fraction of the limit, while large codebases and long sessions use much more. That’s why being able to split research and execution between two products with separate limits is a real practical advantage.

## For whom is it useful

Developers on ChatGPT Pro** are those who already have a subscription with access to GPT-5.5 Pro, but who want to use Codex to write code. Devspace allows you to get the most out of both products at no extra charge.

Commands that actively use Codex – for which the rate limits of Codex become a real limitation. Transferring some tasks (research, planning, code review) to ChatGPT frees up the Codex limit for direct code writing.

Vibecoders working with unfamiliar codebases - when you need to quickly understand someone else's project. GPT-5.5 Pro through devspace can explore the structure and explain the architecture, and Codex then implements the necessary changes.

## Installation and launch

The repository contains .env.example, a standard environment variable template. The main dependence is the presence of a ChatGPT account with access to GPT-5.5 Pro (ChatGPT Pro subscription).

General launch scheme (details in README repository):

# Cloning a repository
git clone https://github.com/Waishnav/devspace.gitX
cd devspace

# Set up environment variables
cp .env.example.env
# Edit .env by specifying the required keys

# Set dependencies and run
npm install
npm run dev

Once launched, the MCP server is available locally – it needs to be connected in ChatGPT settings as an MCP tool. Once connected, ChatGPT accesses the tools described in the server.

## Connection to the MCP ecosystem

devspace is part of a broader trend: ChatGPT supports MCP tools, and developers can build their own MCP servers to integrate external services and local environments with ChatGPT.

This is the way devspace went: instead of waiting for OpenAI to open GPT-5.5 Pro in Codex, the author implemented the desired bundle through the open MCP protocol.

Codex itself continues to actively develop support for MCP: recent updates have added approval choices for MCP actions for the current chat or all chats, improved the marketplace of plugins with separate tabs and filters by category.

## Limitations and honest assessment

devspace is a project built in one day on a hackathon. That means:

What works well: Conceptually and technically, the solution is correct, the task is clear, MCP is the standard protocol supported by ChatGPT.

**What to consider: * The hackathon code usually needs to be reworked before being used in serious projects. The 27 stars at the time of writing are a small but real community.

OpenAI dependency: If OpenAI changes the Separate Rate Limits Policy or adds GPT-5.5 Pro directly to Codex, the main advantage of devspace will partially disappear. Although the MCP connector for local environment exposure in ChatGPT will remain useful in this case.

## Outcome

devspace is an elegant solution to a real problem: GPT-5.5 Pro isn’t available in Codex, but it’s the one that does the best planning and research. The MCP connector opens the local environment for ChatGPT and creates a two-step workflow: scheduling through GPT-5.5 Pro in ChatGPT, execution through Codex with separate rate limits for each.

For developers who are hitting the limits of Codex or want to use a more powerful model for the research part of the work, this is a ready-made tool that you can try right now.

Repository: github.com/Waishnav/devspace