Editing website content through ChatGPT and Claude: an MCP integration example

When people talk about AI on a website, they often mean a chatbot that answers visitors' questions. For one client, we approached it from the owner's side: we connected the website to ChatGPT and Claude so product content could be managed through a conversation.

The aim was practical. Updating a product description should not require copying text between an AI tool and the website administration panel, then repeating the process for another language.

From a visitor chatbot to a tool for the website owner

A customer-facing chatbot and a content management integration solve different problems. One helps visitors find information. The other lets an authorized person work with information stored on the website.

For a manufacturer of outdoor shading products, we built an MCP server into a website with a product catalog. It exposed a defined set of website operations to the connected AI application. The owner could find a product, request a revised description, and have the approved wording saved without manually transferring it into the administration panel.

What a content update looks like

The conversation can begin with an instruction such as:

Find the outdoor blinds product. Break the Czech description into clearer paragraphs and bullet points, then translate the revised text into English. Show me the draft before saving anything.

The AI uses the available tools to find and read the product. The owner can review the proposed wording before asking for the update to be saved. This review step is particularly useful for specifications, product claims, and translations, where fluent wording is not enough to establish accuracy.

Image uploads were also included in this implementation. The owner could attach a photograph and ask for it to be added to the relevant product. These were specific operations provided by the integration, rather than unrestricted access to the website.

A deliberately limited set of tools

The integration was designed around content work. Its available operations included:

• Finding products: searching the catalog by product name or category.
• Reading a product: retrieving the current Czech and English text.
• Updating text: changing a product name or description in one language at a time.
• Uploading photographs: adding images to an existing product.

The tools did not allow the AI to create or delete products, change page URLs, or reorganize categories. Limiting the available operations helped keep the integration focused on the tasks it was intended to support.

Text was sanitized before being stored, and changes were recorded in a history so edits could be traced. These measures address specific risks; they do not mean that every generated sentence will be correct or that an integration can never make a mistake.

Human review still matters

Asking to see a draft before saving is a useful working practice. It should not be confused with a technical approval control unless the application explicitly enforces one. In this workflow, the owner should check the proposed changes and clearly instruct the assistant when to save them.

Working with one language at a time also makes it easier to review what changed. The owner can refine the source description first, then request and check a translation. Product dimensions, compatibility, and other factual details should come from reliable product information rather than an AI guess.

What this changes for content management

The benefit is a shorter path between deciding what needs updating and editing the stored content. The owner can find a product, work on its description, and manage another language in the same conversation instead of repeatedly switching interfaces.

The administration panel remains useful. A chat integration provides another way to carry out selected tasks; it does not need to replace every management screen or grant the AI access to every operation.

Could this work for another catalog or website?

The same approach can be considered for other sites where people regularly maintain structured content. The starting point is to identify the tasks, determine who should have access, and define which operations the connected assistant is allowed to perform.

Connecting management tools to AI is also different from making public content easier for search engines or AI services to understand. An editing integration does not guarantee that an AI service will recommend your business or cite your website.

If you manage a catalog, an e-commerce site, or another content-heavy website, see our custom MCP integration service. Tell us which updates take the most effort and what system holds the content. We can discuss whether a limited set of connected tools would be useful for your workflow.