With AI agents

An AI agent can use roadstyle in two ways: as tools that any MCP app can call, or as a library that the agent writes code against.

Tools, no code: the MCP server

The MCP server lets any MCP-capable AI app (Claude Code, Claude Desktop, Cursor, ...) draw road maps with no code. Each tool saves an HTML map. It returns the path of the map and a PNG preview that the agent can look at.

Add the server to Claude Code:

claude mcp add roadstyle -- uvx roadstyle-mcp

Add it to Claude Desktop (claude_desktop_config.json):

{"mcpServers": {"roadstyle": {"command": "uvx", "args": ["roadstyle-mcp"]}}}

The server saves maps in ~/roadstyle-maps/. The preview needs Chromium once: uvx --from roadstyle-mcp playwright install chromium.

Tool	What it does
render_place(place, network_type="drive", options, output_path)	Downloads the OpenStreetMap roads of a place with osmnx and draws them. network_type is drive, walk, bike, all, drive_service or all_public
render_file(path, options, output_path)	Draws a road file (GeoPackage, GeoJSON, Shapefile, ...)
snapshot(html_path, lon, lat, zoom, bearing, pitch)	Makes a PNG of a saved map. Use it to check street names and arrows: side streets show from zoom 17

options takes any render_edges keyword, for example {"color_by": "maxspeed", "legend": true}. A misspelt keyword gives an error that names the closest valid keyword.

Claude Code plugin: the server and the skill in one install

/plugin marketplace add Khoshkhah/roadstyle
/plugin install roadstyle@roadstyle

Code: the agent skill

An agent that writes Python (Claude Code, Codex, the agent of Cursor) works best with the skill. The skill is one page: the one call, the data contract, the JavaScript API and the traps. To install it for Claude Code:

mkdir -p ~/.claude/skills/roadstyle && curl -fsSL -o ~/.claude/skills/roadstyle/SKILL.md \
  https://raw.githubusercontent.com/Khoshkhah/roadstyle/main/skills/roadstyle/SKILL.md

All the documentation, as a list of links for LLMs: llms.txt.