OpenAI Codex

Set up CE.SDK with OpenAI Codex

OpenAI Codex is a terminal coding agent. Ten minutes of setup turns it into a CE.SDK expert that takes you from an idea like “a business card designer with print-ready export” to a working editor in your browser.

Before you start

You don’t need to be a developer to follow this guide. You need two things on your machine.

  • Codex installed and signed in. Open a terminal and type codex to check.
  • Node.js 20 or newer. It provides the npx command used below and runs the editor you build.

From setup to working editor in four steps

Step 01

Install the CE.SDK skills

Run one command in the project folder you want to build in. It installs the CE.SDK Agent Skills for Codex into .agents/skills/ beside your project: the CE.SDK docs and starter kits for 10 web frameworks, available offline. Add -g to install them once for every project on your machine.

Install the skills for Codex 
npx skills add imgly/agent-skills -a codex
Step 02

Connect the live docs

The MCP server gives Codex real-time search over the current CE.SDK documentation, so answers stay accurate between releases. One command registers it, no config file to edit and no API key.

Add the CE.SDK MCP server 
codex mcp add imgly_docs --url https://mcp.img.ly/mcp

Codex also reads MCP servers from ~/.codex/config.toml, so you can add the same server by hand if you prefer to keep it in the config file.

Or edit ~/.codex/config.toml 
[mcp_servers.imgly_docs]
url = "https://mcp.img.ly/mcp"
Step 03

Verify it works

List the installed skills in your terminal.

Check the install 
npx skills add imgly/agent-skills --list

Then start codex and ask a docs question. Codex picks the right skill from your request; to call one explicitly, prefix its name with $, as in $docs-react. If the answer cites CE.SDK documentation, you’re set.

Ask a test question 
Using the CE.SDK docs, how do I get started with React?
Step 04

Describe what you want to build

Start codex in an empty folder and paste a product idea. Codex scaffolds the project, installs CE.SDK, wires up the features, and prints the command to run it (usually npm run dev). Run it and your editor opens in the browser.

Your first build prompt 
Build a web-to-print business card designer with CE.SDK: professional templates with locked layouts, customer photo upload with background removal, live preview, and print-ready PDF export with bleed margins.

What you can do

Three skills turn Codex into a CE.SDK specialist. Ask in plain language and Codex picks the right one, or call it explicitly with $name.

Build a feature

build sets up a CE.SDK Web project and implements the features you ask for. Describe the outcome and Codex scaffolds and wires it up.

Understand a concept

explain breaks down how CE.SDK works and how the pieces fit together, adapted to your framework and level.

Look up the docs

docs-[framework] pulls reference guides and API docs into your terminal across 10 frameworks, including React, Next.js, and Node.js.

Example prompts

Copy a starting point into Codex. Start a product, extend it feature by feature, or ask how something works before you commit to it.

Scaffold an editor
Build a CE.SDK t-shirt designer with mockup previews, text and clipart tools, and a print-ready export.
Build a CE.SDK ad designer with Unsplash stock photos, background removal, and AI image generation for campaign variants.
Build a CE.SDK photo collage maker with grid layouts and one-click sharing.
Extend it
Add variable data so a CSV of names fills the template.
Add a review step that validates print margins before export.
Lock our brand fonts and colors so users can only edit the text.
Understand it
Explain what CE.SDK gives me out of the box versus what I configure.
Explain the CE.SDK export pipeline and output formats.

Tips

Get better results from the first prompt.

  • Prompt with outcomes, not implementation: “customers personalize a card and download a print file” beats “integrate the export API”.
  • Start in an empty folder so the scaffold doesn’t mix with other files.
  • Re-run the add command anytime to refresh the skills.
  • Skills fire from a plain-language request. To call one explicitly, prefix its name with $, as in $build.
  • Run skills and MCP together: skills for scaffolding, the MCP server for live docs search.

Set up another agent

Point a different coding agent at CE.SDK.

OpenAI Codex

Ship a working editor

Install the skills, paste a prompt, and Codex scaffolds a CE.SDK project you can run.

Read the Agent Skills docs Back to all agents

OpenAI Codex FAQs

No. Codex runs in the terminal, but you steer it in plain language. Describe the product you want and it scaffolds the project and tells you how to run it. You review the result in the browser and keep prompting.

You need an OpenAI account for Codex itself. The CE.SDK skills and MCP server are free and need no API key, and the project Codex scaffolds runs locally without a license key. Licensing only comes in when you deploy what you built.

Both, ideally. Agent Skills bundle the docs and starter kits offline so Codex can scaffold autonomously. The MCP server adds real-time documentation search on top, so answers stay current between releases.

Re-run `npx skills add imgly/agent-skills -a codex` in your project folder and restart the Codex session. `npx skills add imgly/agent-skills --list` shows the available skills. In Codex you call one explicitly with `$name`, for example `$docs-react`.