Let AI agents draw traffic scenes — the drawtonomy MCP server

@drawtonomy/mcp-server is a Model Context Protocol server that lets AI agents draw traffic scenes. Once connected, you can ask your agent things like “a two-lane road with an ego car and a cut-in vehicle ahead” and get a rendered diagram back in the conversation — plus a link that opens the same scene in the drawtonomy editor for hand-editing.

Works with any MCP-compatible client: Claude Desktop, Claude Code, Cursor, VS Code, and others.

Setup

npm install -g @drawtonomy/mcp-server

Then register it with your client. For Claude Desktop, add to claude_desktop_config.json:

{
  "mcpServers": {
    "drawtonomy": {
      "command": "drawtonomy-mcp"
    }
  }
}

For Claude Code:

claude mcp add drawtonomy -- drawtonomy-mcp

Tools

Tool	What it does
generate_scene	Renders a traffic scene from a structured JSON specification as SVG or PNG. The agent writes the JSON from your plain-language description — lanes, vehicles, pedestrians, paths, and text annotations.
open_in_editor	Returns a drawtonomy.com URL that opens the generated scene in the full editor, so you can refine it by hand and export OpenDRIVE / OpenSCENARIO / Lanelet2.

Scene elements follow the editor’s conventions: lane subtypes (road, sidewalk), vehicle templates (sedan, bus, truck, motorcycle, bicycle), and the color language used across drawtonomy examples (ego = blue, threat = red, planned paths = green).

How it fits with the other AI paths

Path	Where it runs	Best for
AI Scene Generator	Inside the editor (Extensions button)	Generating editable shapes directly on your canvas
MCP server (this page)	Your AI client	Quick diagrams in a conversation, handing off to the editor
Headless generation	Your own scripts / CI	Batch generation and format conversion in code

Source

The server lives in the public drawtonomy repository.