aspicio-dxf-viewer
Opens, inspects, and renders DXF/CAD drawings for AI agents: describe_dxf returns structured facts (layers with actual drawn colors, units, bounds, text content) and render_dxf returns PNG images. On MCP Apps hosts like ChatGPT and Claude, view_dxf opens an interactive in-chat viewer with pan, zoom, and layer toggles.
README
<div align="center"> <img src="apps/demo/public/favicon.svg" width="72" height="72" alt="Aspicio logo" /> <h1>Aspicio</h1> <p><strong>DXF understanding for people, applications, and AI agents.</strong></p> <p><em>Aspicio</em> (Latin: "I look at")</p> <p> <a href="https://github.com/frontsail-ai/aspicio/actions/workflows/ci.yml"><img src="https://github.com/frontsail-ai/aspicio/actions/workflows/ci.yml/badge.svg" alt="CI" /></a> <a href="https://www.npmjs.com/package/@aspicio/core"><img src="https://img.shields.io/npm/v/%40aspicio%2Fcore?label=%40aspicio%2Fcore" alt="npm: @aspicio/core" /></a> <a href="https://www.npmjs.com/package/@aspicio/elements"><img src="https://img.shields.io/npm/v/%40aspicio%2Felements?label=%40aspicio%2Felements" alt="npm: @aspicio/elements" /></a> <a href="https://www.npmjs.com/package/@aspicio/react"><img src="https://img.shields.io/npm/v/%40aspicio%2Freact?label=%40aspicio%2Freact" alt="npm: @aspicio/react" /></a> <a href="https://www.npmjs.com/package/@aspicio/vue"><img src="https://img.shields.io/npm/v/%40aspicio%2Fvue?label=%40aspicio%2Fvue" alt="npm: @aspicio/vue" /></a> <a href="https://www.npmjs.com/package/@aspicio/svelte"><img src="https://img.shields.io/npm/v/%40aspicio%2Fsvelte?label=%40aspicio%2Fsvelte" alt="npm: @aspicio/svelte" /></a> <a href="https://www.npmjs.com/package/@aspicio/mcp"><img src="https://img.shields.io/npm/v/%40aspicio%2Fmcp?label=%40aspicio%2Fmcp" alt="npm: @aspicio/mcp" /></a> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="License: MIT" /></a> </p> <p><a href="https://aspicio.frontsail.app"><strong>▶ Live demo</strong></a></p> </div>
Aspicio is an open-source (MIT), TypeScript-first DXF engine: one
framework-free parse → tessellate pipeline that runs in the browser, in
Node, and in Cloudflare Workers. A person gets an interactive WebGL viewer
of a CAD drawing; an AI agent gets structured JSON facts and a rendered
PNG of the same file. Every surface — the browser viewer, the web
components and their React, Vue, and Svelte bindings, the headless
renderer, the HTTP API, and the MCP server — is a thin adapter over the same engine,
so a drawing is equally readable everywhere.
DXF bytes ──parse──▶ DxfDocument ──tessellate──▶ Tessellation ──┬─▶ WebGL renderer (viewer)
(normalized model) (batched geometry) ├─▶ SVG string (export / API / MCP)
└─▶ DrawingSummary (describe)
How it's built: docs/architecture.md · behavior specs: docs/product-specs/
<img src="docs/sample-demo.png" alt="Aspicio viewing a sample floor-plan DXF — layer panel, colored geometry, text, and a dimension" />
Embed it
One embed, every flavor — and every path below renders the same web components, so the result is pixel-identical no matter which you pick.
Web components — plain HTML, any framework
One tag gives you the layer panel plus an interactive preview; no bindings needed:
<script type="module">
import "@aspicio/elements";
</script>
<aspicio-embed src-url="/drawing.dxf" style="height: 480px"></aspicio-embed>
React
The same embed with idiomatic props and a ref exposing the full
viewer, via @aspicio/react:
import { DxfEmbed } from "@aspicio/react";
<DxfEmbed src={file} style={{ height: 480 }} />;
Vue
Typed props and emits with unwrapped payloads, via
@aspicio/vue:
<script setup>
import { DxfEmbed } from "@aspicio/vue";
</script>
<template>
<DxfEmbed src-url="/drawing.dxf" style="height: 480px" />
</template>
Svelte
The same components as raw Svelte 5 source with typed callback props,
via @aspicio/svelte:
<script>
import { DxfEmbed } from "@aspicio/svelte";
</script>
<DxfEmbed srcUrl="/drawing.dxf" style="height: 480px" />
Vanilla TypeScript
Skip the ready-made UI and drive the viewer directly from
@aspicio/core — bring your own chrome:
import { DxfViewer } from "@aspicio/core";
const viewer = new DxfViewer(document.querySelector("#preview")!);
await viewer.load(file); // File | Blob | ArrayBuffer | DXF text (ASCII or binary)
Headless — Node and Workers
Parse, describe, and render with no browser at all (server-side previews, thumbnails, pipelines):
import { parseDxfBytes, tessellate, describeDrawing, tessellationToSvg } from "@aspicio/core";
const doc = parseDxfBytes(bytes); // ASCII or binary DXF
const drawing = tessellate(doc);
const summary = describeDrawing(doc, drawing); // units, bounds, layers, texts…
const svg = tessellationToSvg(drawing);
What you get: WebGL rendering batched to one draw call per layer (large drawings stay interactive), broad entity coverage (lines, arcs, circles, ellipses, polylines with bulges, splines, TEXT/MTEXT, DIMENSION, SOLID/HATCH fills, nested INSERT blocks — anything unsupported is counted and reported, never fatal), a layer list with the colors that are actually drawn (per-entity overrides included, not just the layer table), measure with object snap, entity picking, paper-space layouts, SVG/PNG export, and first-class touch. Out of scope: editing and 3D.
Hand it to an agent
The same engine speaks MCP and HTTP, so an agent can read a drawing instead of guessing at it:
describe_dxf— units, bounds, size, layers with effective colors, entity counts, and the drawing's text content. An agent reads a title block or a dimension value directly — no OCR, no vision round-trip.render_dxf— a PNG of the drawing the model can look at.view_dxf(hosted server) — an interactive in-chat viewer for the person in the conversation, via the open MCP Apps extension: pan, zoom, layer toggles, fullscreen, host light/dark theming. The widget is locked to the drawing the tool call delivered and makes no network requests; hosts without MCP Apps still get the structured facts.
<img src="docs/demo-widget.gif" alt="The in-chat viewer loading a 1.1 MB floor-plan DXF, toggling dimension and text layers, and expanding to fullscreen" />
| Surface | Local files | URLs | Inline DXF |
|---|---|---|---|
stdio MCP — npx -y @aspicio/mcp |
✅ | ✅ | ✅ |
Hosted MCP — aspicio-api.frontsail.app/mcp |
— | ✅ | ✅ |
HTTP API — /describe, /render |
POST body | ✅ | ✅ |
Connect:
- Claude Code — one step installs the MCP server plus the bundled
skills (
aspicio-inspect-dxf,aspicio-embed):/plugin marketplace add frontsail-ai/aspiciothen/plugin install aspicio@aspicio - Codex — the same repo doubles as a Codex marketplace:
codex plugin marketplace add https://github.com/frontsail-ai/aspicio,codex plugin add aspicio@aspicio, thencodex mcp add aspicio -- npx -y @aspicio/mcp - Any client that launches stdio MCP servers — register
npx -y @aspicio/mcp - Any client that supports remote MCP (Streamable HTTP) — point it
at
https://aspicio-api.frontsail.app/mcp(no install; speaks MCP, not a browser page) - Plain HTTP —
GET /describe?src=<dxf-url>,GET /render?src=<dxf-url>&format=png|svg; the API self-describes at/openapi.json
URL fetches are guarded (private-network blocking, size caps, redirect validation, timeouts). The stdio server reads local files in-process and never uploads the DXF to any Aspicio service — though, as with any tool result, your MCP client passes the returned summary or image to its model provider. Full details: privacy policy · terms.
Available today · direction
Everything above is shipped and live: viewer + demo, core, web components, React, Vue, and Svelte packages, headless describe/render, stdio and hosted MCP, the in-chat MCP Apps viewer, the HTTP API with OpenAPI, and plugin packaging for Claude Code and Codex.
Direction (intent, not commitments — see issues): MCP registry listings, structured entity queries and focused rendering, and an upload flow so remote surfaces can handle local files.
Packages
| Package | Description |
|---|---|
@aspicio/core |
The viewer library: parsing, tessellation, rendering, camera, input |
@aspicio/elements |
Web components: <aspicio-embed>, <aspicio-preview>, <aspicio-layer-panel> — plain HTML, Svelte, any framework |
@aspicio/react |
React bindings: <DxfEmbed>, <DxfPreview>, <DxfLayerPanel> |
@aspicio/vue |
Vue 3 bindings: the same three components with typed props and emits |
@aspicio/svelte |
Svelte 5 bindings: the same three components as raw .svelte source |
@aspicio/mcp |
MCP server for AI agents: describe_dxf + render_dxf |
@aspicio/api |
DXF HTTP API server (private): /describe, /render, /mcp |
@aspicio/widget |
MCP Apps in-chat viewer widget (private), served by the api server |
@aspicio/demo |
Standalone demo app (private) — also the reference integration |
How the viewer packages fit together: every framework path funnels into the same Lit web components — one implementation of the embed UI — which sit on the framework-free core. React, Vue, and Svelte get thin veneers with idiomatic props; plain HTML consumes the elements directly.
flowchart TD
REACTAPP["React app"]
HTMLAPP["Plain HTML / vanilla JS app"]
VUEAPP["Vue app"]
SVELTEAPP["Svelte app"]
REACT["<b>@aspicio/react</b><br/><DxfEmbed> · <DxfPreview> · <DxfLayerPanel><br/><i>thin @lit/react veneer, API-stable</i>"]
VUE["<b>@aspicio/vue</b><br/><DxfEmbed> · <DxfPreview> · <DxfLayerPanel><br/><i>thin Vue 3 veneer, typed emits</i>"]
SVELTE["<b>@aspicio/svelte</b><br/><DxfEmbed> · <DxfPreview> · <DxfLayerPanel><br/><i>raw Svelte 5 source, compiled by your bundler</i>"]
ELEMENTS["<b>@aspicio/elements</b><br/><aspicio-embed> · <aspicio-preview> · <aspicio-layer-panel><br/><i>Lit web components — the one embed-UI implementation</i>"]
CORE["<b>@aspicio/core</b><br/>parse → tessellate → render<br/><i>camera · input · picking · SVG/PNG export · headless describe</i>"]
REACTAPP -->|"idiomatic props, ref → DxfViewer"| REACT
REACT -->|"wraps"| ELEMENTS
HTMLAPP -->|"attributes + DOM events"| ELEMENTS
VUEAPP -->|"idiomatic props + emits"| VUE
VUE -->|"wraps"| ELEMENTS
SVELTEAPP -->|"typed callback props"| SVELTE
SVELTE -->|"wraps"| ELEMENTS
ELEMENTS -->|"drives"| CORE
HTMLAPP -.->|"or hand-rolled UI on the DxfViewer API"| CORE
classDef pkg fill:#191c22,stroke:#4c8dff,color:#e7e3da
classDef app fill:#1f232b,stroke:#3a3f4a,color:#9aa0ab
class REACT,VUE,SVELTE,ELEMENTS,CORE pkg
class REACTAPP,HTMLAPP,VUEAPP,SVELTEAPP app
Development
Toolchain: Vite+ (vp) on top of bun.
vp install # install dependencies
vp run dev # start the demo app
vp run ready # check + test + build everything (the repo gate)
Testing, CI/deploy, releasing, and contribution guidance: CONTRIBUTING.md.
Aspicio is developed and maintained by FrontSail AI.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.