Provides the ability to import real street data from OpenStreetMap by geocoding an address and reading street tags, then rendering a cross-section illustration.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@street-generator-mcp Design a residential street with wide sidewalks and trees."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
An MCP server that lets Claude design and render urban street cross-sections from natural language, in the visual style of the Street Generator app, figures and all. It can also read a real street from an address via OpenStreetMap.
Claude does the open-ended reasoning (designing the street). The server does the deterministic work (drawing it).
The server holds no LLM, no API keys, and no secrets. It keeps the model's output small (a
compact StreetConfig instead of thousands of tokens of hand-drawn SVG) and moves the heavy,
exact rendering into code. The result is faster, cheaper, and reliable: every street is drawn
by the same renderer, identical every time.
User: "a calm residential street, wide sidewalks, trees, protected bike lanes"
|
v
Claude -> StreetConfig (JSON), guided by the tool schema
|
|-- render_street(config, style) -> server draws an illustrated SVG
|-- build_share_url(config) -> streetgenerator.com/?s=...
|
v
Claude shows the image and the linkRelated MCP server: agentsumo-mcp
render_street(config, style?)Renders a StreetConfig to an illustrated SVG cross-section (buildings as floor stacks,
sidewalks with pedestrians, planting strips with trees, lanes with cars).
style (optional): colorMode "outline" (default) or "color", plus showLabels,
showMeasurements, showFigures (all default true).
import_street_from_osm(street, houseNumber, city, postcode, country, style?)Reads a real street from OpenStreetMap and renders it. Give a full address (street, house number, city, postcode, country). If any part is missing or unclear, Claude asks first. When several places match, the tool returns lettered candidates (A, B, C) so you can pick one, then it renders the chosen street.
Under the hood: geocode the address (Nominatim), read the nearest street's tags (Overpass),
interpret and map them to a StreetConfig, then render. This tool needs internet access; the
other two do not.
build_share_url(config)Returns a streetgenerator.com link that opens the design in the live app, where you can keep
editing and export it.
Every render returns the cross-section as SVG (which you can save) together with a streetgenerator.com link that opens the same design in the app, where the app's own PNG, SVG, and JSON exports are available. Native PNG output from the server is on the roadmap (see Future work).
From simple to more complex. You type these to Claude; it calls the tools and shows the result.
Prompt | What you get |
"Render a two-lane street with sidewalks on both sides." | A simple SVG cross-section (two traffic lanes, two sidewalks) and a share link. |
"Design a calm residential street with wide sidewalks, street trees, and a protected bike lane on each side." | A fuller illustrated SVG with pedestrians, trees, and buffered cycle lanes, plus a share link. |
"Draw a boulevard with a central median, bus lanes, and buildings, in colour." | A wide coloured cross-section with a median, bus lanes, and building floor stacks. |
"Design a one-way street with parking, and hide the width numbers." | An SVG with labels but no measurements (a style option), plus a share link. |
"Show me the street at Unter den Linden 77, 10117 Berlin, Germany." | The real street read from OpenStreetMap and rendered, plus a share link. |
"Load the street at Hauptstrasse 1, Germany." (ambiguous address) | A lettered list of candidate places (A, B, C) to choose from; pick one and it renders. |
import_street_from_osm queries two free OpenStreetMap services: Nominatim (geocoding) and
Overpass (map data). It is meant for occasional, interactive lookups, not bulk downloading.
Please stay within their public usage policies:
Nominatim: at most 1 request per second, an identifying User-Agent (already set by this
server), results cached where possible, and no automated or bulk querying. See the
Nominatim usage policy.
Overpass: fair use only, roughly up to 10,000 requests and 1 GB of data per day per IP, with rate limiting under load and a 180 second per-request timeout. See the Overpass usage guidance.
For heavy or automated use, run your own Nominatim or Overpass instance instead of the public servers. Map data is copyright OpenStreetMap contributors, available under the ODbL; displayed results should attribute OpenStreetMap accordingly.
{
"mcpServers": {
"street-generator": {
"command": "node",
"args": ["/absolute/path/to/street-generator-mcp/dist/index.js"]
}
}
}Then ask Claude to design or import a street and it calls the tools automatically.
npm install
npm test # unit, mocked-network, and fidelity tests
npm run build # compile to dist/ and copy figure assets
npm run dev # run the server from sourceThe layout math, element data, figure art, and the OSM interpreter/mapper are ported
value-for-value from the Street Generator app (the interpreter and mapper come with their
original unit tests). Fidelity is checked with structural tests in test/.
PNG output from the server (SVG to raster).
validate_street against Berlin RASt rules (the engine already exists in the app).
Nearest-way selection for OSM import (v1 takes the first street within a radius of the geocoded point).
Figures for cycle and bus lanes once art exists.
Render a real street cross-section by reading its layout from OpenStreetMap. Requires a COMPLETE address: street, house number, city, postcode, and country. If the user hasn't provided all of these, or the address is unclear, ASK them for the missing parts before calling. If several places match, the tool returns lettered candidates (A, B, C…); show them to the user, then call again with the chosen candidate's lat and lng.
Inputs: city (required), street (required), country (required), postcode (required), +4 more.
Render a StreetConfig as an illustrated SVG cross-section (people, cars, trees included). Claude produces the StreetConfig from the user's description; this tool draws it deterministically in the Street Generator visual style.
Inputs: config (required), style.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 3 tools
Each tool has a unique purpose: build_share_url creates a sharing link, import_street_from_osm fetches real-world data, and render_street visualizes a config. No overlap in functionality.
All tool names follow a consistent verb_noun pattern with snake_case: build_share_url, import_street_from_osm, render_street.
With only 3 tools, the set is slightly minimal but covers the core workflow of import, render, and share for street generation. The count is reasonable for a focused domain.
The tools cover import, render, and share, but lack any editing or querying capabilities for generated configurations. The user must rely on Claude to produce the StreetConfig, which may be a gap for manual adjustments.
Inspect real parcels, create and edit urban plans, compare alternatives, and share 2D/3D designs.
Design 3D-printable multi-colour map models from OpenStreetMap data, render them, and share them.
OpenStreetMap queries, maps/styles, search, routing, terrain, analysis, pipelines, and rendering.
Text-to-CAD: build exact STEP or 3MF/STL on the server, each stated requirement measured.
