how to use codex in cursor i tried every way to connect them and here is what actually works versus what the internet says works
OpenAI Codex and Cursor are two different AI coding tools and combining them is not as straightforward as most guides suggest. I spent three weeks testing every method for getting Codex capabilities inside my Cursor workflow. Some approaches worked cleanly. Some were deprecated. One popular suggestion I found online no longer works the way it is described. This is the honest current guide to using Codex in or alongside Cursor in 2026.
Marcus Webb
July 8, 2026
Important context before reading: OpenAI Codex has meant two different things at different times. The original Codex was a code-generation model OpenAI released in 2021 that powered early GitHub Copilot. That model was deprecated in March 2023. The new Codex in 2026 is an agentic coding agent accessible via ChatGPT Pro at $200 per month, capable of running tasks autonomously in a sandboxed environment. This guide covers both meanings because search results mix them up constantly.
Method 1: Using OpenAI API Models in Cursor (What Actually Works)
Cursor allows you to configure which AI models power its chat and completion features. In Cursor Settings you can add your own OpenAI API key and select GPT-4o or o3 as the underlying model. This is the legitimate way to get OpenAI model quality inside Cursor. It does not give you the new agentic Codex experience but it gives you the same underlying model intelligence that powers Codex. This method works, is well documented, and is what I use when I want OpenAI model responses in Cursor rather than Anthropic Claude.
# How to Add OpenAI API Key to Cursor
# This gives you GPT-4o quality responses in Cursor chat
## Step 1: Get an OpenAI API key
1. Go to platform.openai.com
2. Sign in or create account
3. Click API keys in left sidebar
4. Create new secret key
5. Copy it immediately (shown only once)
## Step 2: Add key to Cursor
1. Open Cursor
2. Cursor Settings (gear icon bottom left > Cursor Settings)
3. Go to Models tab
4. Find 'OpenAI API Key' field
5. Paste your API key
6. Save
## Step 3: Select your model
1. In Cursor chat (Ctrl+L)
2. Click the model dropdown at the bottom of chat
3. Select gpt-4o, o3-mini, or o3 depending on your needs
4. Your chat now uses that OpenAI model
## Cost of using your own API key:
GPT-4o: approximately $2.50 per million input tokens
o3-mini: approximately $1.10 per million input tokens
For typical Cursor chat usage: $5-15 per month at moderate use
This is separate from your Cursor Pro subscription cost
## Why you might do this:
- Access o3 reasoning model for complex problems
- Use GPT-4o when you prefer it over Claude
- Compare model outputs on the same prompt
- Use a model not yet in Cursor's default listMethod 2: The New Codex Agent Alongside Cursor (How I Actually Use Both)
- What the new Codex agent is: OpenAI's Codex in 2026 is an agentic coding tool accessed through ChatGPT Pro at $200 per month. It runs in a sandboxed cloud environment, reads your repository, executes code, runs tests, and returns results autonomously. It is not a plugin or extension. It is a separate workflow.
- How I use Cursor and Codex together: Cursor is open in my editor all day for active coding, inline completions, and project-context questions. Codex gets specific long-running tasks I delegate to it asynchronously while I continue working in Cursor. Example: I paste in the task, Codex works for several minutes on it, I review the result and bring the changes back into my Cursor project.
- There is no direct integration between them: Codex does not run inside Cursor. There is no Cursor extension that connects to the Codex agent. They are used in parallel workflows. Anyone claiming Codex runs inside Cursor is describing something that does not exist as of 2026.
- Practical workflow: Write feature in Cursor for daily coding. Delegate complex autonomous tasks like writing a full test suite or completing a multi-file refactor to Codex via ChatGPT Pro interface. Review Codex output in your editor (which happens to be Cursor). Apply the changes.
Method 3: The Deprecated Original Codex API (Do Not Try This)
Older guides and some forum posts still reference adding the original Codex model code-davinci-002 or code-cushman-001 to Cursor via an API key. These models were deprecated by OpenAI in March 2023 and are no longer available. If you follow an old guide telling you to add a Codex model endpoint to Cursor you will get an API error. The model does not exist anymore. Do not spend time troubleshooting this. It is not a configuration problem. The model was retired. Use GPT-4o or o3 instead which are substantially better anyway.
What I Actually Got From Three Weeks of Testing
- OpenAI API in Cursor (GPT-4o via API key): works cleanly, costs money per token on top of the Cursor subscription, useful when I specifically want the GPT-4o reasoning style rather than Claude. Took 10 minutes to set up.
- New Codex agent via ChatGPT Pro: powerful for autonomous task completion, not integrated with Cursor, requires switching to the ChatGPT interface. The $200 per month price point makes this a deliberate choice for specific task types rather than daily use.
- Original Codex API: does not work, do not try it.
- The combination I landed on: Cursor Pro as my daily driver with Claude as the default model, OpenAI API key added for tasks where I specifically want GPT-4o or o3, and Codex agent via ChatGPT Pro for autonomous tasks that benefit from its sandboxed execution environment.
Mistakes I Made During Testing
- Trying the deprecated Codex API first because a popular guide from 2022 was the top search result: spent 90 minutes troubleshooting what turned out to be a deprecated model. Always check the date on guides for AI tools. The field changes faster than content gets updated.
- Expecting Codex agent to integrate with Cursor directly: looked for a Cursor extension or plugin that would bring the new Codex agent inside my editor. Does not exist. Accepted the parallel workflow pattern instead.
- Adding my API key without monitoring costs: the first week I added the OpenAI API key I had unlimited chat mode turned on and was casually using it for simple questions. Got a $34 API bill that surprised me. Set a monthly spend limit in the OpenAI platform settings immediately after.
Final Thoughts
Using Codex in Cursor comes down to which Codex you mean. For OpenAI model quality inside Cursor, adding your API key and selecting GPT-4o or o3 is clean and works immediately. For the new agentic Codex experience, it lives in the ChatGPT interface and runs alongside Cursor rather than inside it. The old original Codex model is gone. Once you understand which version of Codex you are after, the path is clear.