Skip to main content

AI generation with Elovate via MCP: what's the difference?

How AI generation through MCP and the AI queue differs from exporting, prompting and re-importing your product data.

There are roughly two ways to enrich product data with AI. The familiar one: run an export, have ChatGPT or another language model rewrite it, and import the result. The newer one: let an AI assistant work directly in your product data through MCP, the Model Context Protocol, connected to Elovate.

Both produce text. The difference is in everything around the generation: how much context the AI has, who checks what goes live, and how mistakes are caught before they reach your webshop. At the centre of that sits the Elovate AI queue.


What MCP is and how it connects to Elovate

MCP is an open standard that lets an AI assistant, such as Claude, work safely with other systems. Elovate has its own MCP server. Through it, the assistant can look into your Elovate environment and work there directly, without any file going back and forth.

Through MCP, the AI can:

  • Read your data: products, categories, attributes, attribute sets, store views and the current value of every field.

  • Measure quality: which fields are empty, where meta descriptions are missing or too long, which products share duplicate texts.

  • Set up workflows: AI actions for attribute extraction, content enrichment and translation, each with its own prompt, model and scope.

  • Test before running: try an action on a few products and judge the result.

  • Propose changes: put improvements in the AI queue as proposals, never as direct edits.

The key rule: nothing done through MCP reaches your webshop on its own. Every change waits for a person in Elovate, and the sync to Magento is always started by a human.


The AI queue: where every AI change waits for a human

The AI queue is the place in Elovate where AI-generated changes land before anything is applied. Whatever the AI produces, it shows up there first, with the current value next to the new one. A person decides; the AI never does.

Changes reach the AI queue in two ways.

1. Direct proposals

For small, targeted changes, the AI proposes an edit to a specific product, category, page, block or blog post. Think of a better meta title, a missing attribute value, a category assignment or a translated attribute label. The AI can also propose a new page or blog post, which is created inactive by default.

  • Every proposal carries a reason the reviewer can judge, such as “meta description was missing” or “description mentioned the wrong material”.

  • An administrator sees the current and proposed value side by side and approves or declines.

  • Only an approval applies the change in Elovate and sends it to the webshop. The sync log shows “AI proposal” as the origin.

  • Proposals can be withdrawn or updated while they are pending. If the product changes in the meantime, the proposal is marked outdated instead of overwriting newer content.

  • An approved change can be reverted: an administrator puts the old values back.

2. Workflow results

For the same kind of change on hundreds of items, a workflow is the right tool. Each generated result becomes a task item that passes two roles:

  • Editor: checks the generated text, corrects it where needed, and approves it or asks for a new generation.

  • Moderator: takes the final decision. Approve and sync, send back to the editor, or decline for good.

Workflows can chain actions, for example attribute extraction, then content enrichment, then translation. A later step only runs for a product once the earlier step's result has been approved by both roles and synced. So a translation is never made from a text nobody has checked.

The two routes at a glance

  • Direct proposal: AI proposal (one item, with a reason) → Administrator approves or declines → Sync to webshop

  • Workflow result: Workflow result (many items at once) → Editor checks and corrects → Moderator approves or declines → Sync to webshop

Either way, nothing reaches the webshop until a person approves it and starts the sync.


The difference in practice

1. The AI sees your whole catalogue, not just a file

With an export, the model only sees the columns you send it. Through MCP, the AI looks up what it needs: the product's category, its attribute specs, the brand, related products and the tone of existing texts. The result is more specific copy that matches your actual product data.

2. Analyse first, then generate

MCP starts by measuring, not writing. The AI shows how complete each field is and where the biggest gains are, for example how many products have no meta description. You enrich where it matters instead of rewriting everything.

3. Workflows instead of one-off prompts

A prompt in ChatGPT is used once. A workflow in Elovate is repeatable: the same prompt, model and scope also run for the new products that arrive next month.

4. A human between the AI and the webshop

Everything the AI produces goes through the AI queue. Nothing is overwritten until someone approves it, and you start the sync yourself. With an import, that check only exists if you organise it.

5. Glossaries, rules and languages are built in

Fixed translations of brand and trade terms, forbidden claims and store-view settings are applied automatically. With an export, you have to put them in the prompt every time.

6. No shifting columns

Anyone who imports knows the risk: a wrong column, a broken delimiter or a SKU that doesn't match. Through MCP the AI writes per field and per product, validated against attribute codes and allowed options.


Side by side

Export → prompt → import

MCP with Elovate

Context for the AI

Only the columns you send

Whole catalogue: attributes, categories, relations, existing texts

Starting point

A file

A quality analysis per field

Repeatability

Prompt again every time

Workflow also runs for new products

Review

Manual, usually spot checks

AI queue: every item reviewed, current vs new value

To the webshop

Straight after import

Only after approval and a manual sync

Languages

A separate round per language

Translation actions per store view, with glossary

Error risk

Shifted columns, encoding, SKU mismatch

Validation on attribute codes and options

Undo

Hard, old values are overwritten

Decline, withdraw, or revert after approval

Barrier to entry

Low, anyone can do it

Elovate account and MCP connection needed


When to choose which

Export, prompt and import works fine for a one-off job on a small set of products, where you read every line yourself. Think of twenty new products for a seasonal campaign.

MCP with Elovate is the better choice once scale, repetition or multiple languages come in. For example when you want to enrich hundreds of products, include new products automatically, or have several people working on the same catalogue.


Conclusion

The difference isn't the AI itself, it's everything around it. With export and import, the AI is a freelance copywriter who only sees what you hand over. Through MCP, the AI works inside your product data, starts with an analysis, and puts every change in the AI queue for a person to approve. You get better copy, less risk and less manual work, and you stay in control of what goes live.

Did this answer your question?