# The vDesign MCP server

> Connect Claude, ChatGPT, Cursor, Claude Code or any MCP app to vDesign, and your assistant makes real, editable designs on a canvas you can keep working on by hand.

vDesign speaks the [Model Context Protocol](https://modelcontextprotocol.io). Connect your
assistant and it can create designs, change them, check them for problems and show you the
result, on the same canvas you edit by hand. The design is not a picture: every text box,
shape and colour stays a layer you can move, retype or recolour.

## Connect in one step

**Remote MCP**, for Claude, ChatGPT and any assistant that takes a connector URL:

```text
https://mcp.getv.design/mcp
```

Add it as a connector in your assistant. The first time, it sends you to a vDesign page that
asks you to sign in and allow the connection. After that, the assistant works on the
documents open in vDesign, in the desktop app or in the [web app](https://getv.design/app/), on any device where
you are signed in. You can see and remove the assistants you have allowed in Settings →
Connect.

**Local MCP**, for Claude Desktop, Cursor, Claude Code and other apps on the same computer as
the desktop app. Turn it on in Settings → Connect, and the app runs a server only that
computer can reach, at `http://127.0.0.1:7823/mcp`. Settings → Connect can write the entry
into Claude Desktop's configuration for you, and copies the path of a small launcher for apps
that start a server from a command.

The [MCP docs](https://getv.design/docs/mcp/) have every detail.

## What your assistant can do

| Tool | What it does |
| --- | --- |
| `vd_create` | Create a document |
| `vd_patch` | Edit a document a few lines at a time, instead of rewriting it |
| `vd_read`, `vd_list` | Read a document, list the open ones |
| `vd_show` | Show the design in the chat |
| `vd_problems` | Check a document for anything that did not draw as asked |
| `vd_skill` | Learn the vDesign format |
| `vd_add_icon`, `vd_search_icon` | Find icons and place them as real vector paths |
| `vd_add_asset`, `vd_assets`, `vd_remove_asset` | Add, list and remove images and music |
| `vd_create_brand`, `vd_read_brand`, `vd_apply_brand` | Make a brand and put it on a design |
| `vd_open` | Open a document in vDesign (remote) |
| `vd_export`, `vd_frames` | Export a document, and see a video's key moments (desktop) |

An assistant needs no instructions from you to use them. The server tells it to learn the
format first and to build anything large a region at a time, so you watch the design arrive
rather than wait for it.

## Brands and video, from the chat

Ask your assistant to make a brand from your website or the text of your guidelines, and
vDesign's code reads the exact colours, fonts and logo while the assistant writes the voice and
rules. Every design it makes after that names the brand's roles, and "switch this deck to our
other brand" is one step. See [a brand kit](https://getv.design/make/brand).

A design can also play: ask for a short social video and the assistant writes the motion, words
rising, numbers counting, a camera moving between scenes, as part of the same file. See
[a social video](https://getv.design/make/video).

## Why a design tool built for agents

Most design tools that offer MCP were built for a mouse first. Their documents are large JSON
trees, HTML, or a format the app keeps to itself, so an assistant either rewrites a lot to
change a little or works through a narrow set of commands.

A vDesign document is a short text file written for a model to read and write: one line per
element, geometry as relationships (`x=(@logo.right + $gap)`), colours and sizes as named
tokens. To change the spacing of a whole layout, the assistant writes one line,
`set $gap = 40`, and everything that reads `$gap` moves. That is what makes edits by an
assistant fast and cheap, and why the canvas fills in as the reply streams. See
[the .vd format](https://getv.design/format).

## Not only through your assistant

The same agent is built into the app as **vDesign AI**, on your own key: Anthropic, Google,
OpenAI, xAI, Mistral, DeepSeek, OpenRouter and others, or a local or custom model in the
desktop app. Use whichever you prefer for a design, or both.

## Questions

### Which assistants work with the vDesign MCP server?

Any assistant that takes an MCP connector URL can use the remote server at https://mcp.getv.design/mcp, including Claude and ChatGPT. Apps on the same computer as the vDesign desktop app, such as Claude Desktop, Cursor and Claude Code, can use the local server instead.

### Does it cost anything?

vDesign is free for individuals. Your assistant's own plan is whatever it already is; vDesign adds no charge.

### Can the assistant read files on my disk?

No. It works on the documents open in vDesign, never on files on your disk. It cannot read a document you did not open, and saving stays something you do.

### Do I need to teach the assistant the format?

No. The server tells the assistant to load the vDesign skill before its first write, either through the MCP Skills extension or through the vd_skill tool.

### What can it make?

Anything people look at, on one canvas, such as posters, slides, brochures, social posts, menus, brand guidelines, one-page websites and short videos with motion and music.