How to integrate Codex models into Claude Code? (Complete tutorial included)

recent Claude Another wave of account bans has begun, and my main account was not spared.

Getting your account banned is a minor issue, but... Claude The code contains many memory settings and scheduled tasks, which cannot be disabled...

See OpenAI Tibo, the core product manager, shared a usage tip. I followed the instructions and successfully converted Codex's gpt-5.6-sol into an API for integration within minutes. Claude In the code.

Now you can use up your Codex credit limit. Claude You can continue using it in the code, and you don't have to worry about your account being banned. It's really great.

This article provides a detailed guide to the steps, so even beginners can get started.

Claude Code cannot directly call Codex's OAuth interface. We can perform authentication and protocol conversion locally using the CLIProxyAPI, and then use CC Switch to write the connection configuration, which will allow us to... Claude Calling models from Codex from Code.

Install CLIProxyAPI

No need to manually download and configure from GitHub, just open Codex and add the following...Prompt wordsSend to Codex:

Prompt wordsPlease help me install and configure CLIProxyAPI on Windows.

Official project address: https://github.com/router-for-me/CLIProxyAPI
Require:
1. Download only from the project's official GitHub Releases.up to dateVersion;
2. Choose the Windows_amd64 or Windows_arm64 installation package based on your computer's architecture;
3. Install to the CLIProxyAPI folder in the current user's directory;
4. Copy config.example.yaml to config.yaml;
5. Bind the service to 127.0.0.1, using port 8317;
6. Disable remote management and generate a random management password and downstream password. API Key；
7. Retain the default auth-dir; do not read, copy, or display OAuth Tokens.
8. When you need to log in to your Codex account, pause the operation, tell me the command to execute, and I will complete the login in the browser;
9. After logging in, start the service and request /healthz and /v1/models to check if it was successful;
10. Finally, tell me the location of the Base URL, API Key in the configuration file, and the list of currently available models.

Please execute the command directly.automaticCompleted steps. When encountering errors, first check the logs and try to fix them; do not just provide operation instructions.

Codex willautomaticHelp us install the CLIProxyAPI and list key information such as connection details and available models.

Account Login

If you are prompted to log in during the installation process, you can complete the login by running the following command in PowerShell:

.\cli-proxy-api.exe –config .\config.yaml –codex-login

Start the service and verify that the API is working correctly in the new PowerShell:

.\cli-proxy-api.exe –config .\config.yaml

Create a new PowerShell window and run the following script to query available models:

$match = Get-Content .\config.yaml | Select-String '^\s*-\s*"([^"]+)"\s*$' | Select-Object -First 1 $key = $match.Matches[0].Groups[1].Value.Trim() $result = Invoke-RestMethod ` -Uri "http://127.0.0.1:8317/v1/models" ` -Headers @{ Authorization = "Bearer $key" } $result.data | Select-Object id

If the console returns a list of models, such as gpt-5.6-sol, it means that the local proxy bridge has been successfully built!

Configure CC Switch

We open CC Switch and select... Claude After selecting the code, click the "+" in the upper right corner to add a supplier.

Supplier Name → CLIProxyAPI Codex

API Key → Query and copy from the information provided by Codex

Base URL→http://127.0.0.1:8317

API Format Selection: Anthropic Messages

Do not enable "Local Routing" and "Full URL Mode"

The model mapping can be filled in as needed, or you can fill in all of them with gpt-5.6-sol.

Once the configuration is complete, we can return to the CC Switch main interface and click "Test Connection," which will show that the connection is normal.

Now we open it in PowerShell Claude Once you have the code, you can directly use gpt-5.6-sol.

After successful integration, I immediately set up several commonly used scenarios to see how gpt-5.6-sol works in Codex. Claude After coding, how well do you perform?

Image generation

We are Claude Enter the requirements in the code:

Prompt words`/generate-image` creates an infographic.SimpleThe easy-to-understand illustrations demonstrate five basic yoga poses and provide instructions on body postures and breathing techniques.

Claude The code successfully understood the task and invoked the backend gpt-image-2 via the /generate-image workflow to generate an infographic:

The layout is very pleasing to the eye, the images and text are very clear, the Chinese rendering is correct, and the instructions are understood with great accuracy.

We put the samePrompt wordsTry sending it to Codex:

The generated infographics, both in layout and content, are consistent with... Claude The generated code is very close, indicating that the image generation capability invoked by the tool is not diminished after the proxy conversion.

Chat and edit videos

Let's try Clsude Code's MCP plugin again.automaticTransformation capabilities.

I chose ChatCut. Claude Enter the following in the Code field:

Prompt wordsRead chatcut.io/claude, help me install the ChatCut plugin and create a new task.

Claude Code will read the installation instructions and configure the plugin.

Midway through the process, the browser will pop up a login page, requiring us to manually confirm once.After installation, create a new one Claude In the Code session, ChatCut will...automaticload。

We create a new one Claude Code dialogue allows you to communicate through dialogue. AI They helped us edit the video.

Prompt wordsCreate a new task using ChatCut. Use D:\00\materials\material1.mp4 to material20.mp4 as video footage and D:\00\materials\Don't Let Love Wither Violin.mp4 as background music to create a 9:16, 1080×1920, H.264 vertical beat-synced video.

Edit 20 clips into a 9:16 vertical video. Select only one best continuous segment from each clip, and ensure each segment appears at most once in the final video. Arrange the clips in order based on the content and rhythm of the music; file numbering is not required.

Analyze the rhythm, accents, melodic shifts, and emotional changes of the background music, and edit accordingly. Switch scenes at strong beats, rhythmic changes, and melodic transitions; appropriately extend shots during gentler parts of the music, and accelerate scene transitions during stronger rhythms and climaxes.

The editing primarily uses clean, crisp cuts, with natural transitions used sparingly where necessary, avoiding overuse of transitions and effects. The background music fades in naturally at the beginning and fades out naturally at the end.

The first time it is executed,AI The project was successfully created and all materials were uploaded, but due to the complexity of the task, ChatCut only uploaded the materials and did not provide the complete editing steps.

I broke down the instructions so that ChatCut could execute them step by step.

First, adjust the canvas to portrait mode and add an audio track, then add fade-in and fade-out effects to the music.

Prompt wordsContinue using ChatCut's official MCP to edit the current project. Do not create a new project, re-upload materials, or export.

Please modify the current Main Timeline:

1. 1. Change the canvas to 9:16 portrait mode, resolution 1080×1920, and maintain 30fps.
2. Create audio track A1.
3. Place the entire 27.43-second audio file, "Don't Let Love Wither Violin.mp4", into A1, starting from the 0th second.
4. Add a natural fade-in at the beginning of the music and a natural fade-out at the end.
5. Do not add any visual assets in this step.

Once finished, check and report the canvas size, audio track, music start time, and total timeline duration; do not export.

This time it was completed successfully. ChatCut added the correct audio track to the canvas, added fade-in and fade-out effects, and adjusted the canvas size.

We continued with the editing tasks:

Prompt wordsContinue editing the current project and the current Main Timeline; do not create a new project or re-upload materials.

Add all materials from material 1.mp4 to material 20.mp4 to the V1 video track.

Editing requirements:

1. Each selected material can only appear once and cannot be reused.
2. Unsuitable materials can be omitted; do not force them to be added just to reach a certain quantity.
3. Only one best continuous segment is selected from each submitted material.
4. Use enough high-quality clips to cover the 27.43-second background music, and there should be no gaps in the timeline.
5. If the number of selected clips is small, you can extend the quality clips appropriately, but avoid repeating scenes.
6. Turn off the original audio for all video footage, keeping only the background music in A1.
7. useintelligentCropping should be adapted to 1080×1920 portrait screens, keeping the main subject intact and avoiding stretching, squashing, or distorting the image.
8. Arrange the order and cuts of the images according to the main downbeats, melodic shifts, and emotional changes in the music.
9. The music is used in slightly longer shots during the soothing parts, and the transitions between the increased tempo and the climax are accelerated.
10. The style primarily uses clean, sharp cuts, employing only a few necessary natural transitions.
11. Do not add subtitles, titles, logos, watermarks, or complex effects.

After confirming that everything is correct, export the video in 9:16, 1080×1920, 30fps, H.264 format, and return the project link and video download link.

ChatCut automaticPerform audio accent analysis, select material segments, sort, trim, and export.automaticHere's the link to download the full movie!

Let's take a look at the final cut:

The standard 9:16 portrait mode features no repetitive visuals, clean and crisp transitions, and natural fade-in and fade-out music.

ChatCut prioritized matching light drum beats when identifying beats, so the sense of rhythm wasn't as strong as a professional editor's, but the overall effect of the finished product was still pretty good. If it were out of 100, I would give this task 80 points.

Coding

Daily Use Claude The code requirements will be quite extensive, so let's test it out.

Prompt wordsPlease implement an interactive 3D smart city digital twin screen that runs within a browser in the current empty repository.

Technology stack:

– React + TypeScript + Vite

– Three.js or React Three Fiber

– Post-processing effects can use existing libraries

– Not dependent on paid resources or external APIs

Visual effects:

– Dark Future City

– A large number of buildings of varying heights

– Roads, illuminated traffic flow, streetlights, and dynamic billboards

– Random illumination from building windows

– Ground with grid pattern and slight reflection

– Bloom glow, fog effects, and day/night cycle

– Opening animation

- Maintain smoothness and avoid unnecessarily increasing the model's face count.

Interactive features:

1. Rotate, zoom, and pan the scene using the mouse.
2. After searching for a building, it will be highlighted and a building information panel will pop up.
3. Search the building and let the camera...automaticFly over.
4. As day turns into night, the sun, sky, and city lights change in sync.
5. Simulate vehicles moving along a road.
6. The left side displays indicators for population, energy consumption, traffic, and air quality.
7. The chart data is constantly changing dynamically.
8. Click the Traffic, Energy Consumption, and Security buttons to switch between different visualization layers.
9. The traffic layer displays the road heat map effect.
10. The energy consumption layer changes color according to the building's energy consumption.
11. The security layer displays camera scan cones and alarm points.
12. Supports pausing animation and resetting camera.
13. Correctly adapts when the page size changes.
14. Added a performance panel and image quality options.

The city must be generated procedurally, using a fixed random seed to ensure reproducibility. Please implement, run, and start the development server directly. Ensure that the complete city and animation are displayed on the first launch; do not use static screenshots to impersonate a 3D scene.

Let's take a look at the final result:

Prompt wordsThe 3D city, which requires real-time rendering, has been completed. Buildings, roads, traffic, and streetlights have all been rendered, and we can adjust the viewpoint and scene using the mouse.

After the day-night cycle, the sky, building lights, and urban atmosphere will also change synchronously, resulting in a relatively complete overall effect.

Searching for buildings will highlight them and pop up an information panel. The overall effect is very good.

Daily Workflow

Besides writing code, we can try to delegate any daily tasks that involve SOPs to others. AI Go do it.

For example, my download folder is often overflowing with various files, which makes... AI Organized by type and time:

Prompt wordsPlease help me organize the folder D:\Browser Downloads. The steps are as follows:

1. First, scan the files and generate an archiving scheme based on type and date. No folder processing is required.
2. Write a Python script to organize files into a "type/year/month" directory.
3. Before proceeding, show me the files that will be moved and wait for my confirmation.
4. After confirmation, perform the cleanup without deleting or overwriting any files.
5. After completing the process, check the results and generate an operation log for easy undoing.

Please complete the task directly; don't just tell me how to do it.

Let's take a look. Claude Code provides the following organization solutions:

I categorized all my files by file type, year, and month.

We choose to confirm.Claude The code has just started executing; it categorizes all files:

Give Claude Code changed GPT The model was originally just something I wanted as a temporary fix, but I didn't expect it to work so well. Previously... Claude Terminal habits, historical context, and environment configurations in the code can all be preserved, so you don't have to adapt to a new set of tools.

Previously we chose AI Tools are often equivalent to the model behind them. Now, with Claude Code serves as a work platform, allowing users to freely choose models and continuously reuse their accumulated workflows.

Of course, different models are not entirely compatible with complex tool calls and context compression. Before implementing them in important projects, it is still recommended to run a few real-world tasks for testing.

Model manufacturers and Agent The client will also gradually decouple, and in the future, users may not choose a fixed model, but a system that allows different models to truly get the job done.

Models can be changed, tools can be changed, but the workflows you've accumulated don't need to be completely overhauled and rebuilt.

Original link:Claude After being blocked, I managed to connect gpt-5.6-sol in a few minutes. Claude Code