Claude can write your positioning statement, name your company, and argue with you about your target market for an hour. Then you ask it for a logo and the conversation stops. You get a description of a logo. You open a new tab, start over in a design tool, and retype everything Claude already knew.
The Zoviz MCP server closes that gap. Connect it once, describe a business in a sentence, and you get back a real brand identity in the same conversation: logo concepts, a colour palette, typography, social assets, and production files in SVG, EPS and PDF.
This guide covers what it does, how to connect it, and what it costs.
What is an MCP server?
The Model Context Protocol is an open standard that lets an AI assistant call outside tools during a conversation. Without it, Claude can only describe what a tool would do. With it, Claude can actually use the tool and hand you the result.
Zoviz now speaks that protocol. Once connected, Claude reaches your Zoviz workspace directly, and the brand you create lives in your account rather than disappearing when the chat ends.
How to connect Zoviz to Claude
Setup takes about two minutes and works the same way in Claude Desktop and on claude.ai.
Step 1. Open Claude, go to Settings, then Connectors.
Step 2. Choose Add custom connector. Name it Zoviz. In the MCP server URL field, enter:
https://mcp.zoviz.com/mcpStep 3. Click Add, then Connect, and sign in with your Zoviz account.
That is it. Claude will now list the Zoviz tools as available, and you can start asking for brand work in plain language.
The server also works with Cursor, Windsurf, Perplexity, OpenClaw, Hermes and NemoClaw. Because your brand lives in the Zoviz workspace rather than inside one assistant's memory, it follows you between them. Generate an identity in Claude on Monday, pull the same files into Cursor on Wednesday, and it is the same brand rather than a regenerated approximation of it.
What the Zoviz MCP server can do
Fifteen operations are available once connected.
Building a brand add_brand creates a brand from a description. create_logo generates logo concepts from it. generate_more_logos explores further directions when the first set is not right. get_brand_list and get_brand_logo_files retrieve anything you have made before.
Working with images generate_image creates original imagery. image_edit refines it by prompt. remove_background isolates a subject. upload_media brings your own files into the workspace.
Producing marketing assets generate_social_content builds platform sized posts and stories from the brand you just created, so the palette and typography carry through instead of drifting.
Files and account download_logo returns production files. get_credits, get_account_info, get_workspace_memberships and media_confirm handle the plumbing.
Shows the exact prompt structure and settings for shooting a cinematic AI short video with no camera, then swapping the entire scene while keeping the subject and camera move locked in place.
→ How to Make and Edit a High-Quality Short Video with Seedance
A worked example
Here is a real sequence, start to finish, in one conversation.
You: Start a brand for a Nordic cold brew coffee company. Minimal, dark green, aimed at people who work from home.
Claude calls add_brand, then create_logo. You get logo directions back in the chat with the palette and type choices attached.
You: The second one is closest, but the mark feels too heavy. Try some lighter variations.
Claude calls generate_more_logos, constrained to the direction you picked.
You: Good. Now generate the Instagram post set for a launch week announcement.
Claude calls generate_social_content. The posts come back in the same palette and typeface as the logo, because the brand is a persistent object that every later call reads from.
You: Give me the vector files.
Claude calls download_logo. You get SVG.
The thing worth noticing is what did not happen. You never re briefed. You never opened a dashboard. The colour that was chosen in step one was still correct in step four.
Walks through turning one AI-generated brand kit into a full beauty brand commercial, all built in a single day.
→ From AI Brand Kit to AI Commercial: How I Built a Beauty Brand in One Day
How this compares to other design MCP servers
Zoviz is not the first creative MCP server, and it is worth being straight about that. Figma ships one. Canva ships one. Onbrand, Frontify, zeroheight and Paper.design all have entries in the category, along with several open source logo generation servers on the public directories.
The difference is what sits at each end of the pipe.
Most design MCP servers assume a design already exists. Figma's connects an assistant to files you have already built, with access to frames, components and design tokens, which is exactly right if you are a product team turning components into code. Frontify and zeroheight expose brand assets and documentation that already live in their portals. Those are retrieval problems, and they are solved well.
Most design MCP servers assume a design already exists. The difference is what sits at each end of the pipe.
Zoviz starts from nothing. There is no prior document, no template library to work inside, no canvas to open first. A sentence goes in, and a brand identity system comes out.
The sharpest single difference is the export list. Canva's MCP documentation lists PDF, PNG, JPG, PPTX and MP4. Zoviz returns SVG, EPS and PDF alongside the raster formats. That matters more than it sounds: a sign maker, an embroiderer or a print shop will reject a PNG, and vector is the difference between owning a logo and owning a picture of one.
A real case study on shooting an entire short film with AI in a single day, with no set, no camera, and no crew involved.
→ The Short Film That Needed No Set, No Camera, No Crew. One Day, with Zoviz
What it costs
You can connect the MCP server and create a brand on a free Zoviz account, and the logo will be saved to your dashboard. The first trial is free.
Downloading files and running generation require a subscription. Any paid plan includes full MCP access, and there is no difference between the plans in what the MCP server can do, so pick on the basis of the rest of Zoviz rather than on MCP features.
AI content generation runs on credits, drawn according to what you ask for. Heavier requests, such as long image sets or video work, consume more than a single logo does.
Three things that were harder than expected
If you are building an MCP server yourself, these are the parts that took the longest.
Consistency does not survive independent tool calls.
The naive design has the assistant call a logo generator, then later call a social post generator, and you get two designs that share nothing. That is why add_brand exists as its own step rather than being folded into create_logo. The brand has to be a persistent server side object that everything downstream reads from, or the output drifts within a single session.
Chat interfaces make waiting look like failure.
Generation takes real time. Silence in a conversation reads as a crash, so people re prompt, and now you have two divergent sets of results and double the load. Progress reporting mattered more to the finished product than the generation pipeline did.
"Give me the logo" is underspecified.
It is unambiguous to a person and meaningless to a server. Which brand, which concept, which variant, which format, at what size. Most of the tool schema design went into making the failure mode a clarifying question rather than the wrong file arriving silently.
Try it
Add https://mcp.zoviz.com/mcp as a custom connector in Claude and ask it to build a brand for something. It takes two minutes, and a free account is enough to see whether the output is any good.
If it breaks, tell us. That is more useful to us right now than praise.
FAQ
How do I add an MCP server to Claude?
Settings, then Connectors, then Add custom connector, then paste the server URL and connect. For Zoviz the URL is https://mcp.zoviz.com/mcp. The same three steps work for any remote MCP server.
What is the difference between Claude Skills and MCP?
A Skill is a set of instructions that changes how Claude approaches a task using what it already has. An MCP server gives Claude a genuinely new capability by connecting it to an outside system. Zoviz is the second kind: Claude cannot produce a vector logo file on its own, and the MCP server is what makes that possible.
What data can Claude access through the Zoviz connector?
Your Zoviz workspace, scoped to the account you sign in with. That covers the brands you create, the assets in them, and your credit balance. It does not reach anything else on your machine.
Can I use the logos commercially?
Check the terms attached to your plan before committing a mark to production, and do your own trademark search in the jurisdictions you trade in. No generation tool, ours included, performs a trademark clearance for you.
Does it work with anything other than Claude?
Yes. Cursor, Windsurf, Perplexity, OpenClaw, Hermes and NemoClaw are supported.
