🛠️ text-to-cad：2.8k Stars 的 AI Agent CAD 技能包，一句话生成工业级 3D 模型
搞硬件设计的人都知道，从想法到 STEP 文件的流程有多痛苦。开 Fusion 360、拉尺寸、画草图、倒角、导出 — 一套下来半小时起步。但如果你已经在用 Claude Code 或 Codex 写代码了，那 text-to-cad 就是给这些 coding agent 装上 CAD 大脑的东西。
text-to-cad 不是又一个 3D 建模软件。它是一个 agent skills 合集，让你直接用自然语言告诉 AI：「做一个 100×60×20mm 的方块，上面打 4 个 8mm 通孔」，AI 就能自动生成可生产的 STEP/STL/3MF 文件，还能在 CAD Explorer 里预览。
🛠️ text-to-cad：2.8k Stars AI Agent Skills for CAD — Generate Production-Ready 3D Models from One Sentence
Anyone who's done hardware design knows the pain of going from idea to STEP file. Open Fusion 360, pull dimensions, sketch, chamfer, export — at least 30 minutes. But if you're already using Claude Code or Codex, text-to-cad is what gives those coding agents a CAD brain.
text-to-cad isn't another 3D modeling app. It's a collection of agent skills that lets you tell your AI in plain language: "Make a 100×60×20mm block with four 8mm through-holes" — and it generates production-ready STEP/STL/3MF files, viewable in CAD Explorer.
---
## 项目概况
- **Stars：** 2,860+ / **Forks：** 342
- **语言：** JavaScript（技能脚本）+ Python（build123d 核心）
- **许可证：** MIT，作者 [earthtojake](https://github.com/earthtojake)
- **主页：** [cadskills.xyz](https://www.cadskills.xyz)
- **创建：** 2026 年 4 月，不到一个月冲上 2.8k⭐
## Project Stats
- **Stars:** 2,860+ / **Forks:** 342
- **Language:** JavaScript (skill scripts) + Python (build123d core)
- **License:** MIT, by [earthtojake](https://github.com/earthtojake)
- **Homepage:** [cadskills.xyz](https://www.cadskills.xyz)
- **Created:** April 2026 — hit 2.8k⭐ in under a month
---
## 核心技能拆解
text-to-cad 不是单一工具，是 7 个技能的打包合集，覆盖从建模到制造的完整链路：
🔥 **CAD Skill** — 核心。基于 build123d（Python）生成参数化 3D 模型，输出 STEP、STL、3MF、DXF、GLB。最骚的是它支持 `@cad[...]` 引用机制 — 你改了一段代码，AI 不需要从头生成整个模型，只需要增量更新引用部分。
⚡ **step.parts Skill** — 直接从 step.parts 目录搜索和下载标准件：螺丝、螺母、轴承、垫片、电机、连接器。不用自己画，直接引用现成的工业标准件。
🛠 **URDF / SDF / SRDF Skills** — 机器人描述文件生成。URDF 定义机器人连杆和关节，SDF 适配 Gazebo 仿真，SRDF 配置 MoveIt2 运动规划。一句话生成完整的机器人模型 XML + mesh 引用。
🔧 **CAD Explorer Skill** — 本地跑的 3D 预览工具。生成的文件直接在浏览器里查看，支持 orbit、缩放、截面。
🎯 **SendCutSend Skill** — 直接对接 SendCutSend 工厂。检查材料厚度、SKU 兼容性、二次加工可行性，确保你生成的文件能直接下单生产。
## Core Skills Breakdown
text-to-cad bundles 7 agent skills covering the entire design-to-manufacturing pipeline:
🔥 **CAD Skill** — The core. Generates parametric 3D models via build123d (Python), exports STEP, STL, 3MF, DXF, GLB. The killer feature: `@cad[...]` reference handles — modify source code and AI only needs to incrementally update the referenced geometry, not regenerate everything.
⚡ **step.parts Skill** — Search and download standard mechanical components from step.parts: screws, nuts, bearings, standoffs, motors, connectors. No need to model common parts from scratch.
🛠 **URDF / SDF / SRDF Skills** — Robot description file generation. URDF for robot links/joints, SDF for Gazebo simulation, SRDF for MoveIt2 motion planning. Generate complete robot model XML with mesh references from a single prompt.
🔧 **CAD Explorer Skill** — Local 3D preview tool. View generated models in-browser with orbit, zoom, and cross-section.
🎯 **SendCutSend Skill** — Direct integration with SendCutSend manufacturing. Validates material thickness, SKU compatibility, and secondary operations so your generated files are production-ready.
---
## 实操：5 分钟跑起来
安装就两行：
```bash
git clone https://github.com/earthtojake/text-to-cad.git
cd text-to-cad
```
装到 Claude Code：
```bash
./scripts/claude-install.sh
```
装到 Codex：
```bash
./scripts/codex-install.sh
```
然后用自然语言告诉你的 agent：
> "Create a centered 100 x 60 x 20 mm block with four 8 mm vertical through-holes. Add a 2 mm chamfer on the top outer perimeter."
AI 会生成 Python 脚本（基于 build123d），然后自动输出 STEP 文件。你可以直接在 CAD Explorer 里 orbit 查看效果：
```bash
# 启动 CAD Explorer
./scripts/explorer-start.sh
# 或者用 npx 一行搞定（不需要克隆仓库）
npx agent-skills-cli add earthtojake/text-to-cad
```
## Quick Start: 5 Minutes to Your First CAD
Install is two commands:
```bash
git clone https://github.com/earthtojake/text-to-cad.git
cd text-to-cad
```
Install for Claude Code:
```bash
./scripts/claude-install.sh
```
Install for Codex:
```bash
./scripts/codex-install.sh
```
Then tell your agent in plain English:
> "Create a centered 100 x 60 x 20 mm block with four 8 mm vertical through-holes. Add a 2 mm chamfer on the top outer perimeter."
The AI generates a Python script (build123d under the hood) and exports STEP files automatically. View results in CAD Explorer:
```bash
# Start CAD Explorer
./scripts/explorer-start.sh
# Or one-liner via npx (no clone needed)
npx agent-skills-cli add earthtojake/text-to-cad
```
---
## 对比 & 适合谁
这不是帮你学 CAD 的 — 这是给**已经在用 coding agent 的开发者**准备的。如果你：
- 搞机械设计、机器人、3D 打印，日常被 STEP 导出搞烦
- 用 Claude Code / Codex 写代码，希望 AI 也能直接出 3D 模型
- 做机器人开发，需要快速生成 URDF 做仿真
那 text-to-cad 是目前最顺手的方案。相比之下，传统 CAD 参数化需要手写 OpenSCAD，text-to-cad 用 Python + build123d，对开发者友好太多了。
唯一注意：本地需要装 Python 3.11+ 和 OpenCascade 依赖（`pip install build123d` 解决），不是纯浏览器方案。
## Who It's For
This isn't for learning CAD — it's for **developers who already use coding agents**. If you:
- Do mechanical design, robotics, or 3D printing and hate the STEP export treadmill
- Use Claude Code / Codex and want AI to output 3D models directly
- Build robots and need quick URDF generation for simulation
text-to-cad is the most developer-friendly option. Compared to OpenSCAD's custom syntax, Python + build123d is a massive DX improvement.
One caveat: you need Python 3.11+ and OpenCascade locally (`pip install build123d`). Not a pure browser solution.
---
## 要点总结
- text-to-cad 是 7 个 agent skills 的合集，让 AI coding agent 直接生成工业级 3D 模型
- 核心 CAD Skill 用 build123d（Python），输出 STEP/STL/3MF/URDF 等标准格式，支持 `@cad[...]` 增量引用
- 安装一行命令，用自然语言描述零件，AI 自动生成可生产的文件
- 2026 年 4 月上线，不到一个月 2.8k stars，342 forks，增速很快
- 适合已经用 coding agent 的开发者做机械设计、机器人、3D 打印
## Key Takeaways
- text-to-cad bundles 7 agent skills that turn AI coding agents into CAD generators
- Core CAD Skill uses build123d (Python), exports STEP/STL/3MF/URDF with `@cad[...]` incremental references
- One-command install, describe parts in natural language, AI generates production-ready files
- Launched April 2026 — hit 2.8k stars and 342 forks in under a month
- Best for developers already using coding agents who do mechanical design, robotics, or 3D printing