The image features a blue printer with a pink screen on top of it, sitting on a gray base. The printer is positioned towards the left side of the frame and appears to be ready for use. There are two other objects in the scene - one located at the bottom right corner and another near the center-left area.

How a B2B Integration Software Company
Connected ChatGPT to Drupal

AI Readiness & Enablement

WebDev

Executive Summary

The marketing team was already building website pages with ChatGPT. It had two custom GPTs, one for landing pages and one for blog posts, each loaded with HTML templates. The GPTs had no access to the Drupal CMS, so the team transferred every page manually, and generated pages could contain hardcoded styles.

Darwin gave ChatGPT a direct connection to Drupal through a Model Context Protocol (MCP) server. The assistant can now look up existing pages, assemble a page from the site's Gutenberg blocks and save it as an unpublished draft. The team can continue refining the draft in ChatGPT, with Drupal handling the final review and publication.

The workflow has been tested on a real webinar page, created through ChatGPT and saved directly in Drupal. This provides a working foundation for using AI in the company's website content production process.

The Approach

Darwin structured the work through the Darwin Flux stages, beginning with an audit of the company's website and Drupal setup.

Surface

Audit of 6,920 content records, 14 content types and 80 Gutenberg components, plus the page templates the company's editors used most.

Connections

MCP server built into Drupal with 17 content tools for ChatGPT. Each user logs in with their own Drupal account.

Clarity

Every page is validated and saved as a draft. The assistant can read published pages but cannot change or publish them.

Momentum

Editors can keep refining a draft in ChatGPT or in Drupal. The same connection now helps the team plan updates to page components.

Introduction

The client is a B2B integration software company that helps businesses manage complex supply chains and data flows.

Its marketing team needed to create and update landing pages, webinar pages, case studies and blog posts quickly. ChatGPT was already part of the workflow. The goal was for generated pages to arrive in Drupal in a form editors could continue working with.

The Challenge

The custom GPTs worked from a content brief and a few HTML templates. They had no information about the company's fields, blocks, media library or forms.

Pages Built Outside the CMS

The GPTs produced standalone HTML with hardcoded styles. A page could look fine in the browser and still have nothing in common with the blocks the rest of the site used.

Design Drift on Landing Pages

Of 723 landing pages, 172 used style tags with no Gutenberg blocks, and 39 mixed the two. Pages styled this way do not pick up changes to shared blocks, so fonts, colors and layouts can slowly diverge. The team wanted a consistent site and also preferred the more modern look of its newer HTML pages over the older designs.

Manual Transfer Into Drupal

Each generated page was transferred into Drupal manually. Forms, images and links had to be selected again, and some were reused from unrelated campaigns.

Pages That Were Hard to Edit

A page that looked acceptable in the browser could still be difficult to edit in Drupal's editor, since its layout lived in pasted HTML.

The Solution

Darwin built a custom integration that allows ChatGPT to create and edit pages in the company's Drupal CMS. The integration was built around the website's existing architecture, custom content components and editorial workflow.

Auditing the Drupal Website

Darwin examined the Drupal codebase, custom Gutenberg blocks, content types, theme, media library and forms. Unstable remote access led the team to create a local copy of the website, including its database and 33 GB media library.

The audit covered 6,920 content records, including 6,475 published, and 80 active blocks. A review of page templates and revision histories identified reusable layouts and custom styling requirements that the AI assistant would need to follow.

Building the Initial ChatGPT Skill

The first version used a custom ChatGPT skill with references to the company's content blocks and page-building rules. It generated page files that a Drupal importer converted into unpublished drafts, ready for review and editing.Darwin tested this workflow using landing page briefs and a case-study format.

Improving Page Layout Accuracy

Initial tests exposed a gap between technically valid Drupal content and the intended page design. Some generated pages used the correct blocks but produced layouts that differed from the company's references.Darwin refined the skill using 14 reviewed layouts covering 12 landing page types and a case-study format. Testing also revealed specific issues with page length, hero sections and media selection. An EDI landing page, for example, was shorter than its reference and used an unsuitable hero block.The team introduced additional layout variants and image selection rules to improve the accuracy of generated pages.

Connecting ChatGPT Directly to Drupal Through MCP

Darwin developed an MCP server and a custom CMS skill to give ChatGPT direct access to the company's website content and functionality. The integration provides 17 tools for finding pages, reading block schemas, selecting media and forms, validating content, managing drafts, generating previews and uploading images.The MCP server initially ran as a separate Node.js service. Darwin later rebuilt it as a PHP module within Drupal, bringing the integration onto the company's existing hosting infrastructure.

Securing Content Access and Publishing

Darwin implemented OAuth authentication using the company's existing Drupal user permissions. The assistant can reference published content and manage unpublished drafts. Publishing and changes to live pages remain restricted.Additional safeguards prevent duplicate page creation and reject updates based on outdated draft versions, protecting content from conflicting changes.

Validating the Integration

Darwin first tested the integration on a local copy of the company's Drupal website, where a connected AI agent created and edited pages using the site's existing blocks and layouts. The integration was then deployed to a test site. The company's website marketing lead connected it to ChatGPT from his own account and created pages from real content.

Want AI-created pages your editors can change in the CMS? Darwin integrates your AI tools with your CMS and keeps every page on your site's blocks.

Key Outcomes

ChatGPT Connected to Drupal Through MCP

ChatGPT can work directly with the company's CMS, using existing page layouts, content blocks, media and forms to produce pages consistent with the website's design.

About 2 Minutes per Page

During a live call, the team created three pages, each generated in about two minutes.

Editing in ChatGPT or Drupal

The marketing team can request revisions in ChatGPT or edit drafts in Drupal. The assistant can retrieve the latest changes from the CMS.

Publishing Remains Under Team Control

The integration respects existing Drupal permissions. Pages are saved as unpublished drafts for the team to review and publish.

Before-and-After
Comparison

Metric

Page creation
CMS integration
Website context
Editing

Before

Standalone HTML from custom GPTs
Manual content transfer to Drupal
Raw HTML with inline CSS
Manual edits in Drupal

After

Drupal drafts built using existing site blocks
Pages saved directly through ChatGPT
Access to existing pages, blocks, media and forms
Revisions through ChatGPT or Drupal

Conclusion

Darwin delivered and validated a working ChatGPT–Drupal integration that brings the company's website components and content workflows into the AI assistant. The marketing team tested the integration with real content, demonstrating that page creation and editing can take place directly in ChatGPT while Drupal retains control over publishing.

The project also creates opportunities to extend the integration as the website evolves, including a review of existing components with the client's designer to support newer page designs and a potential rollout to preproduction.