VectorCraft MCP Server: Let AI Agents Draw Vector Art

MCP needs the desktop app or the CLI. The browser editor on this site can’t be driven by an agent; for prompt-based artwork in the browser, use the AI SVG generator.

Two ways to run it

Mode	How	What works
Headless	`vectorcraft-cli mcp --headless`	An in-process engine with a CPU renderer. Everything except the tools that need the window (`inspect_ui`, `type_text`, `open_panel`, window screenshots).
Remote	`vectorcraft-cli mcp --connect 127.0.0.1:7979`, with the app started as `vectorcraft --control 7979`	Everything. Tool calls go to the running app, so you watch the document change live.

With no flag, the server tries 127.0.0.1:7979 and falls back to headless.

Set up with Claude Code

Get vectorcraft-cli: on macOS it is the separate vectorcraft-cli-0.9.0-macos-universal.zip release download; the Linux .deb and .rpm install it as /usr/bin/vectorcraft-cli; the Windows installer puts vectorcraft-cli.exe in C:\Program Files\VectorCraft\. Then register it, confined to one working folder:

claude mcp add vectorcraft -- /path/to/vectorcraft-cli mcp \
  --automation-read-root /work/project \
  --automation-write-root /work/project

Or add it to .mcp.json in your project:

{
  "mcpServers": {
    "vectorcraft": {
      "command": "/abs/path/to/vectorcraft-cli",
      "args": ["mcp",
        "--automation-read-root", "/work/project",
        "--automation-write-root", "/work/project/out"]
    }
  }
}

The two root flags imply headless mode. To confine a live app instead, start it with the same flags: vectorcraft --control 7979 --automation-read-root <dir> --automation-write-root <dir>, then connect as usual.

What the agent gets

Drawing: draw_path (points or SVG path data), draw_shape (rectangle, ellipse, polygon, star, line), set_paint, add_text (point, area or path type), transform, pathfinder, apply_effect, create_graph.
Files: open_file (every format VectorCraft reads), save_file, export (SVG, PDF, EPS, DXF, PNG, JPEG, WebP, layered PSD and more), screenshot for a PNG the model can look at.
Everything else: list_commands and run_command reach every engine command, and command_batch runs several in order, each step able to use an earlier step’s result.
Ready-made prompts: poster, icon-set, recolor, trace-and-style and export-set.

Prompts to try

“Draw a set of six weather icons on a 24-point grid with 2-point strokes and export each as SVG to out/.”
“Open logo.png, trace it with the Flat Logo preset, expand it and save it as logo.svg.”
“Recolour poster.svg to this palette: #1a1a2e, #e93d82, #f5f5f5. Then render a preview.”

Without MCP: the CLI on its own

# Open, recolour and export in one command
vectorcraft-cli run --in in.svg \
  --cmd select.all --cmd paint.setFill --params '{"color":"#ff0000"}' \
  --export out.svg --export out.png --scale 2

# The command catalogue as JSON
vectorcraft-cli commands

FAQ

Can an AI agent control the browser version on this site?

No. Browsers can’t listen on a network port, so the web build has no control channel. The MCP server is part of vectorcraft-cli and drives either its own headless engine or the desktop app. In the browser, use the AI panel for prompt-based SVG and image generation.

Which agents work with it?

Any MCP client that can launch a local stdio server: Claude Code, Claude Desktop, Cursor, VS Code agents and others. The project documents the Claude Code command; other clients take the same command and arguments in their JSON config.

Can the agent read or write files anywhere on my disk?

Not if you confine it. With --automation-read-root and --automation-write-root, every file the server reads must be inside the read folder and every file it writes inside the write folder; a root you leave out grants none of that access. Without either flag, file access is not restricted, so give the flags for real projects.

Source: VectorCraft’s MCP server documentation and control protocol. Flags can change between releases; check vectorcraft-cli mcp --help.