Once the Frontify MCP server is connected, you work with your brand by asking for what you need in plain language. This article covers what to ask for, how to phrase it so you get a useful answer, and where the limits are.

This is for anyone using the MCP day to day: brand and creative teams, marketers, content managers, agency partners, and designers or developers working in tools like Cursor and Figma Make. If you haven't connected yet, start with Frontify MCP for activation and setup.

This is the single biggest source of confusion, so it's worth understanding before you write your first prompt.

The MCP connection isn't scoped to a hub, a brand, or a library. When you authenticate, the AI tool can see everything your Frontify user account can see. If you have access to eight brands, the agent starts every conversation with all eight in view and no idea which one you mean.

On top of that, the MCP retrieves content through Frontify's Content API, which doesn't yet support semantic search or navigation hierarchy. The agent can't browse your guidelines the way a person would, and it can't infer that "the summer campaign" lives in a particular project.

The practical effect: vague prompts produce vague results. The agent may search the wrong brand, return assets from an old campaign, or tell you it can't find something that exists. None of this means the connection is broken. It means the agent needs the context you'd give a new colleague on their first day.

Four habits fix most of it.

Tip: Tell the agent which brand you're working in at the start of a conversation. It will usually carry that context through the rest of the thread, so you don't have to repeat it in every prompt.

Because there's no semantic search, the agent matches on names that exist rather than on what you meant. If you're unsure of a name, ask for the list first and then act on what comes back.

This works for brands, libraries, projects, guidelines, templates, color palettes, and workflow statuses. It takes one extra turn and prevents the most common failure, which is an agent reporting that something doesn't exist when it's really sitting under a name you didn't guess.

Each section below shows a vague prompt and a version that works. Adapt the specifics to your own brand and library names.

Use cases 1 to 3 are read-only, so there's nothing to undo. Most teams should start there.

Find approved assets without opening Frontify or asking someone where things live. The agent searches across brands, libraries, and projects, and returns asset details including preview and download links.

Vague: "Do we have any videos from last year's summer campaign?"

Better: "In the Acme brand, search the Video library for assets tagged summer campaign created between May and September 2025. List the title, file type, and whether each one is approved for external use."

Vague: "Find me the approved logos for the EMEA market."

Better: "In the Acme brand, look in the Logos library and list every logo variant with EMEA in its metadata or targets. For each one, give me the file name, format, and download link."

Ask to see the assets, not just read about them. If you end your prompt with a request to display the results, supporting AI tools render the assets inline as previews with their metadata, alongside options to open them in Frontify or download them directly. Without that instruction you'll usually get a plain text list.

Confirm usage rights before you use anything. Search results can surface copyright and license information, and assets that need license consent may require approval before they can be downloaded. Ask for this explicitly rather than assuming a result is cleared for use.

Ask questions about your guidelines and get answers from the live source rather than a PDF someone exported last quarter. The agent reads guideline page content, and AI tools with vision capabilities can also analyze images and diagrams inside guidelines, so specs that live in a visual aren't missed.

Vague: "What does our tone of voice say about B2B writing?"

Better: "Open the Acme brand guidelines, find the tone of voice section, and summarize the guidance that applies to B2B audiences. Quote the relevant rules and tell me which page they came from."

Vague: "Where can I find our brand colors and approved fonts?"

Better: "For the Acme brand, retrieve the brand colors and fonts. Give me the color names with hex values, and the font names with the weights we're licensed to use."

If you're building in a code or design tool, ask for the output in the format you'll actually use.

Asking for the source page is worth doing every time. It lets you check the answer, and it tells you whether the agent found the real guidance or inferred something.

Use your guidelines to review work that already exists. This is the fastest way to catch off-brand copy or incorrect logo usage before anything is published, and because it's read-only, nothing in Frontify changes.

Guidelines often contain more than tone descriptions. If your brand includes preferred terminology, a do and don't list, or a brand dictionary, the agent can check a draft against those specific rules rather than offering a general opinion.

Vague: "Does this sound like us?"

Better: "Here's a draft product page for the Acme brand. Retrieve our tone of voice guidance and our preferred terminology, then review the draft against both. List every line that conflicts, quote the rule it breaks, and suggest a replacement. Don't change anything in Frontify."

Vague: "Check if this logo usage is right."

Better: "Look at this packaging mockup. Retrieve the Acme logo usage rules from our brand guidelines, including clear space and minimum size, and tell me whether this usage complies. Cite the guideline page for each point you make."

Ask for the rule alongside the verdict. A flagged line you can trace back to a guideline page is something you can act on, or push back on.

Hand the agent a batch of new files and have it upload them to the right place with the right structure. This needs a write-enabled pack.

Vague: "Can you upload these and organize them?"

Better: "Upload these 12 files to the Acme brand, Campaign Assets library, into a new folder called Spring 2026 Agency Delivery. Set the title from the file name, add the tags spring-2026 and agency-supplied to each one, and list what you uploaded when you're done."

Ask the agent to read the result back from Frontify. A summary tells you what the agent believes it sent. Reading the values back tells you what's actually stored. The difference matters when a field didn't apply or a file landed in the wrong folder.

Enrich metadata across many assets at once, including alt text, descriptions, and custom metadata fields. This is one of the clearest time savings the MCP offers, and one of the easiest to get wrong at scale.

Vague: "Can you add alt text to all the product images in this folder?"

Better: "In the Acme brand, Product Imagery library, find the assets in the Footwear SS26 folder. For each one, look at the image and write alt text of 125 characters or fewer describing the product and its setting. Use the existing file name and description as context. Show me the proposed alt text for all of them before you write anything to Frontify."

That last sentence matters. Ask to review the proposed values first, then approve the write, then ask the agent to read the applied values back. A bulk operation applied to the wrong folder is tedious to reverse.

Generate finished designs from your brand templates with custom text and imagery, without opening Figma or the Frontify editor. The agent picks the template, fills in the variables, exports the file, and can save it back to Frontify.

Vague: "I need social banners for our spring sale in German, French, and Italian."

Better: "List the brand templates available in the Acme brand. Use the Social Banner 1080x1080 template to create three versions of our spring sale banner with the headline and call to action in German, French, and Italian. Keep the layout and imagery identical across all three. Export each as PNG and upload them to the Campaign Assets library in a folder called Spring Sale Social."

Ask for the template list first if you're not sure of the exact name. Guessing at a template name is a common reason the agent comes back empty-handed.

Route new assets through review by creating workflow tasks, adding checklist items, and assigning assets to the right stage.

Vague: "These photos need to go through review."

Better: "Create a workflow task called Q2 Product Photography Review in the Acme brand and assign the 14 assets in the Q2 Product Shoot folder to it. Add a checklist covering color accuracy, logo placement, model release on file, and alt text completed. Assign it to the brand review stage."

Name the assets and the stage explicitly. The agent has no way to know which review process you mean if your account has several.

If you don't know what the stages are called, ask the agent to list the available workflow statuses first, then tell it which one you want. Phrases like "send it for approval" only work reliably once the agent can see the status names and match your intent to one of them.

Chain several steps into one request. These are the most useful prompts and the most demanding, because every assumption the agent makes compounds through the chain.

Vague: "Submit this campaign brief and find assets that match."

Better: "Here's our Q3 partner campaign brief. Working in the Acme brand only: 1) Summarize the brief in five bullets. 2) Find approved logos suitable for partner co-branding. 3) Find imagery in the Campaign Assets library tagged partner or B2B. 4) Pull the relevant sections of our co-branding guidelines. Present it as a table with asset name, type, location, and why it fits. Don't upload or change anything."

Vague: "Put together a collection for the partner toolkit."

Better: "In the Acme brand, create a collection called Partner Toolkit 2026. Add the partner lockup logos from the Logos library, the co-branded templates from the Templates library, and any imagery in Campaign Assets tagged partner-approved. Show me the list of assets you plan to add before you create the collection."

Number your steps for multi-part requests. It gives you a checkpoint for each stage and makes it obvious where things went sideways if the output isn't right.

Replace Acme and the library names with your own.

If the MCP isn't returning what you expect, or you want help scoping which tool pack fits your team, contact your customer success manager or Frontify Support.