.png)
Subscribe to our newsletter
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
.svg)
.svg)
.svg)
I recently connected Codex to a local Optimizely CMS 13 project through Model Context Protocol (MCP) to explore how it could work with CMS content. I used it to inspect content types, create an unpublished page, and assemble a Visual Builder experience using existing components.
In my original walkthrough on Optimizely World, I covered the configuration and prompts. Here, I’ll expand on where I see practical uses, the permissions issue I encountered, and what I would test before using this approach on a live website.
MCP allows an AI application to interact with tools and information exposed by another system. In this setup, it gave Codex access to content and metadata in my CMS through VS Code.
I started by asking Codex to retrieve the available content types and inspect their properties. I then asked it to create a draft page and a Visual Builder experience using existing sections and components. Both were saved as drafts without publishing.
I see potential in everyday content management tasks, such as finding and updating pages, creating new content, and building Visual Builder experiences from existing components.
Optimizely Graph APIs could also provide a way to search and retrieve CMS content through AI tools. That is another possibility to explore beyond the tasks covered in this test.
For a team trying this approach, I’d suggest starting with content searches or creating simple draft pages and experiences, then reviewing the results in CMS before publishing.
The setup was fairly straightforward. I installed the Optimizely MCP package in my local CMS 13 project, configured the MCP endpoint and client-secret authentication, and connected Codex through VS Code.
One issue I encountered was permissions. Although Codex connected successfully, it could not create content until I explicitly granted the MCP client access in CMS.
A successful connection does not necessarily mean the client has permission to create or update content. The appropriate CMS access rights still need to be configured.
My first priority would be permissions: making sure the AI tool can perform only the actions it is allowed to take.
I would also test how reliably it creates and updates content, especially with Visual Builder, and how it handles errors or unexpected requests.
For a live website, I would keep content changes as drafts for editors to review and approve before publishing, rather than allowing AI to publish directly.
This local test demonstrated that Codex could work with content types, pages, and Visual Builder through MCP. Before taking that into a live environment, I would focus on validating the intended tasks and keeping editorial review in place.
Read my original walkthrough on Optimizely World here.
Senior .NET Backend Developer
Mardy Abdelkareem is a Senior .NET Backend Developer at First Line Software with over 10 years of experience in backend development, system integrations, and CMS modernization.
He is a certified Optimizely CMS PaaS, CMS SaaS, and Opal Tools Developer, with hands-on experience in Optimizely implementations and CMS migrations.
He is also an active contributor to the Optimizely community, where he shares technical articles, explores Optimizely CMS 13, Graph, and AI-assisted development through hands-on experiments, and helps fellow developers through community discussions.
Last Updated: October 2026