Icecat Studio AI Assistant & MCP Growth: Icecat Studio – Sprint 107 Release Notes

Sprint 107 was built around the main release: the Icecat Studio AI Assistant. Alongside it, the new Story Builder reached our dev environment, the MCP Server got more capable. Several smaller improvements, including a simpler aggregation architecture, round out the sprint.

Introducing the Icecat Studio AI Assistant

Imagine an editor you can simply talk to, on any page of Studio. Give it a product link or a file, and it returns a finished story. Then keep chatting to refine it. Now it’s in Studio and you can find it sitting next to your profile.

Always at hand

• Opens on any page: an AI Assistant button in the header opens the chat panel wherever you are.
• Aware of where you are: it sees your current page context. The empty chat screen shows scenario cards for that page, split into Actions and How-to.
• “What can I do here”: A location tooltip shows where the assistant is open and, upon hover, what it can do there compared to other pages.
• More room to work: the chat window can open full-page, the left sidebar is larger and can expand further, and the prompt input is bigger and resizes automatically.

Finding and creating assets

• Search: finds assets by brand, MPN or name and shows their stories and statuses.
• Create: creates assets from an MPN or ASIN.

Creating stories

• Any source: builds a story from a website link, an attached file (XML, JSON, CSV, HTML), pasted text or images.
• It reads and reasons: it reads the product page, looks at the images, picks modules from the layout and assembles the story itself.
• Respects your wishes: instructions like “no carousels” or “no size block” are followed.
• Self-correcting: it fixes small issues such as mobile crops, alternating sides, and text/video pairs before creating the story.
• Transparent about gaps: it tells you what from the source page did not make it into the story.
• When a page can’t be opened: if a page is protected or behind a login, the assistant asks for a file or text instead.

Working on the story afterwards

• It keeps context: it remembers the source page, its own plan and what is already done, so you can continue in the same chat without explaining everything again.
• Edit content: change texts.
• Edit design: change module styles (background, spacing, borders, rounding).
• Edit structure: add, remove and reorder modules, including adding a module from a source page.
• Edit layout: it works with the story layout as well as individual modules.

Publishing and help

• Statuses and publication: moves stories to approved and runs publication. It can also explain where a story is published and why it is not.
• Answers questions: how-to questions are answered from the Studio documentation.

Safe by design

• Confirmation first: every data change is shown on a confirmation card, and nothing happens until you confirm.
• Respects permissions: it works within your permissions. A user limited to certain brands, languages or categories only sees and changes those.

Verified on real content
We checked about 100 stories from websites, JSON, XML and HTML files, images, chat text and several languages. Almost all were created on the first attempt. On the same pages, the stories are fuller, and the text is taken verbatim.

Story Builder – Central Section Is on Dev

The native Story Builder is now deployed on our dev environment and is being actively tested.

• Modern design: every element on the canvas is more interactive.
• Structure tree: elements are also reachable from the tree as a sub-level.

MCP Server – SVG Import in Progress

We started adding an SVG import tool to the MCP Server, so connected AI agents can upload SVG files. This is the first step in a set of MCP additions planned for the next sprint.

Home Page Revision

The concept for the first page after login was discussed and reworked this sprint actively. The goal is a page that informs everyone who enters the platform. Data and structure come first, and design follows.

Under the Hood: Simpler Aggregation

Aggregation was previously handled by a separate microservice that tracked events and made selections. We simplified this and removed the service and now aggregated stories live in an entity. The collection stays separate, but we process it on assets and entities separately.

Also in This Sprint

• Search filters: “Campaign name” and “Company” in Asset Search. Searching by Company also finds exclusive and targeted story variants.
• Languages alignment: all 81 locale names in Studio now match the language names used in Icecat Brand Cloud. Technical codes are unchanged.
• Asset History: work on a more detailed record of the actions taken on an asset continues.
• Other fixes: UI and API fixes and refinements across the platform.

What’s Next

• Story Builder – Central Section: testing of the version deployed on dev.
• Home Page revision: agree on the data and structure first and move to final design.
• MCP Server: finish the SVG import tool, add a copy-product tool shared with the AI Assistant, and support for Studio AI Assist tools.
• Story v1 access: HTML packages of v1 stories become downloadable through Studio endpoints. Until now, only decomposed v2 stories were published there, and v1 stories are custom HTML pages that cannot be decomposed.