---
格式版本: 2
标题: "Onshape is bracing for an AI explosion"
原文链接: "https://www.engineering.com/onshape-is-bracing-for-an-ai-explosion/"
发布日期: "2026-08-19"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:configured_publication_date_rule"
发布时间证据: "engineering-search-publication-date html:original: <time datetime=\"2026-08-19\""
发布时间校准原因: "信源发布日期识别规则直接确认发布时间"
发布时间校准置信度: "high"
发布时间候选数量: 1
发布时间严格候选数量: 1
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-08-20T02:23:28+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-08-20T02:22:12+08:00"
入库时间: "2026-08-19T18:23:28.581Z"
来源平台: "Engineering.com 搜索"
搜索渠道: "source_template"
搜索词: "https://www.engineering.com/?s=AI%20Rack"
匹配关键词:
  - "AI Rack"
  - "AI"
相关厂家:
  []
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 5
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "该链接为Engineering.com搜索入口页，正文实为Onshape CAD软件AI功能报道，未涉及超节点、AI Rack、机柜级AI基础设施、供电液冷或互连等任何相关技术或商业内容，与项目主题明显无关。"
AI质检模型: "ali-deepseek-v4-flash"
AI质检时间: "2026-08-20T02:24:38+08:00"
AI主题相关性: 0
AI来源权威性: 5
AI新颖性: 0
AI技术细节: 0
AI商业部署信号: 0
AI完整性: 5
AI摘要: "Onshape通过Onshape Labs发布FeatureScript MCP服务器，让用户借助Claude、ChatGPT等AI编码代理创建自定义CAD工具。"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-07T03:16:22.521Z"
采集批次: "2026年8月20日0点36分19秒"
采集批次ID: "20260820-003619-923"
去重键: "https://www.engineering.com/onshape-is-bracing-for-an-ai-explosion"
---

With the new FeatureScript MCP server, anyone’s AI coding agent can create custom CAD tools.

This is Engineering Paper, and here’s the latest design and simulation software news.

First, follow-up: [Last week](https://www.engineering.com/the-agentic-era-for-mechanical-design/) I wrote about Leo AI, a mechanical engineering copilot that CEO Maor Farid claims can create designs indistinguishable from those of a human engineer. In other words, he said, Leo passes the “mechanical Turing test.”

Farid has now elaborated on that idea in an op-ed for Engineering.com: [Opinion: AI has passed the mechanical Turing test](https://www.engineering.com/opinion-ai-has-passed-the-mechanical-turing-test/). Check it out to learn more about Leo and Farid’s philosophizing on AI-assisted engineering.

On to today’s top story.

## Onshape opens up to AI with FeatureScript MCP server

Onshape’s newly launched early-access program, Onshape Labs, has an exciting new experiment: the [FeatureScript MCP (model context protocol) server](https://www.ptc.com/en/news/2026/onshape-launches-featurescript-mcp-server).

An MCP server itself is not that exciting. Onshape is [hardly](https://www.engineering.com/autodesk-announces-fusion-mcp-servers-and-more-ai-updates/) the [first](https://www.engineering.com/now-sketchup-has-an-mcp-server-for-claude-based-3d-modeling/) CAD platform to [adopt the protocol](https://www.engineering.com/siemens-and-xometry-to-bring-instant-quoting-dfm-analysis-to-designcenter/), which opens an application to AI platforms like Claude, Gemini, or ChatGPT. What is exciting is that FeatureScript, Onshape’s custom programming language, hitherto useful but often out-of-reach, might suddenly become the star of the whole platform.

“A lot of our users are just not developers to the degree necessary to build a more elaborate custom feature,” Cody Armstrong, senior director of Onshape AI innovation, told me. Now, with the FeatureScript MCP server, those users can set their favorite chatbot to the task instead.

“You connect your AI to our MCP server toolkit,” Armstrong said. “It’s purpose-built for not only creating custom features, but testing and evaluating those features as well.”

![](https://www.engineering.com/wp-content/uploads/2026/08/SCR-20260618-iuwn-scaled.png)

With the new FeatureScript MCP server, Onshape users can use AI agents to develop custom features like this spec-driven flywheel sizer. (Image: Onshape.)

What kind of features are we talking about? Theoretically, since Onshape itself is built in FeatureScript (every change to every CAD model is captured as a line of viewable FeatureScript code), the sky’s the limit. But the immediate value is in creating custom features to automate repetitive parts of your workflow.

Armstrong and Darren Henry, senior vice president of general operations at PTC (which owns Onshape), showed me several examples during a recent demo. They ranged from simple features, like one that generated a bottle thread on a selected face, to more sophisticated, like one to generate configurable sprockets. Henry showed me a feature that visualized the fill level of a tank and another that calculates how an arbitrary chunk of ice would float in fresh water—or salt. Armstrong even showed me an entire [shell-and-tube heat exchanger](https://gamma.app/docs/Shell-and-Tube-Heat-Exchanger-Parametric-Design-qccnddbyzmp3z8m) built with FeatureScript.

“I don’t know how to program,” Henry emphasized. “Everything I showed you was 100% AI generated.”

Since MCP is an open protocol, the FeatureScript MCP server works with any compatible AI platform. Henry says a coding LLM works best. He prefers Claude Code, but told me others on the team use ChatGPT’s Codex or GitHub Copilot.

![](https://www.engineering.com/wp-content/uploads/2026/08/onshape_ai-mcp-server_claude-code_hxe_2026-05-26-2-scaled.png)

A shell-and-tube heat exchanger feature built with FeatureScript. (Image: Onshape.)

I’ve [commented before](https://event.on24.com/wcc/r/5352236/B052B76915B60340CF7EDEA1D8CBF151) that AI coding agents could be an interesting shortcut to AI-based CAD, since generating code seems to be much easier than generating geometry. Regular readers may remember [HelloTriangle](https://www.engineering.com/hellotriangle-turns-python-code-into-simulation-ready-meshes/), a Python-based CAD tool with an [AI coding agent](https://www.engineering.com/siemens-and-xometry-to-bring-instant-quoting-dfm-analysis-to-designcenter/), as one of several examples of this approach.

But Onshape is the most established of all of them, and Henry believes the platform is onto something big.

“We’re excited because we think this takes CAD in a brand new direction,” he told me. “You need a production level CAD system that’s powerful, but now we’ve opened the gate so it can be customized to your needs. We think these next few weeks we’re going to see thousands of examples.”

If you’re an Onshape user of any subscription tier, you can access the FeatureScript MCP server [here](https://cad.onshape.com/appstore/apps/Onshape%20Labs/6a29aea7c03f8bf659841734). Let me know what you think of it at [\[email protected\]](https://www.engineering.com/cdn-cgi/l/email-protection#b3ded2dfd1d2f3d2c1c1dcc4d5dfca9dd0dcde).

## SimScale releases AI agent for simulation

Onshape isn’t the only cloud-based engineering software releasing AI features.

SimScale has [launched an AI agent](https://www.simscale.com/press/engineering-ai-agent-open-to-community/) for its cloud simulation platform that, according to the company, “can run complete simulation workflows autonomously, from CAD geometry clean-up, to simulation set-up, through to validated output report.”

In a video accompanying the announcement, SimScale flashes example prompts including “Can you help me respond to an RFQ?”, “Please set up the crush test” and “Let’s look at design alternatives using Physics AI models.”

![](https://www.engineering.com/wp-content/uploads/2026/08/SimScale-Community-AI-Agent-2.jpg)

SimScale’s new AI agent sets up crush tests. (Image: SimScale.)

SimScale says the AI agent has been “proven with enterprise customers” over the past year.

“For most of engineering history, simulation was something a handful of specialists did at the very end of a project, long after the decisions that mattered had already been made,” said SimScale CEO David Heiny in the company’s press release. “What 900,000 users tell us is that this has flipped. AI-native engineers are testing ideas on day one.”

You can also watch Heiny in conversation with Engineering.com’s Jim Anderton in [AI in Engineering: Separating Hype from Reality](https://www.engineering.com/a-i-in-engineering-separating-hype-from-reality/).

## One last link

From Engineering.com senior editor Michael Ouellette: [Scaling AI agents is where things get complicated](https://www.engineering.com/scaling-ai-agents-is-where-things-get-complicated/).

Got news, tips, comments, or complaints? Reach me at [\[email protected\]](https://www.engineering.com/cdn-cgi/l/email-protection#a0cdc1ccc2c1e0c1d2d2cfd7c6ccd98ec3cfcd).