cadAI

Can cadAI Let an AI Agent Build Models Through MCP?

Yes. The MCP interface is built for AI agents: an agent drives modeling directly through it, calling the five connected intents, S-01 primitives, S-02 holes, S-03 conversion, S-04 measurement, and S-05 preview. The result is a draft, and a human previews and confirms it before export. Keep two layers apart. MCP is the developer channel, where an agent calls intents from a program. The one sentence assistant in the web version is the other layer, for people who do not write code, and it calls the same five intents.

cadAI is a browser-based parametric CAD from Aiglade, for mechanical part modeling, hole drilling, measurement, and export, with the same kernel on desktop, in the browser, and over MCP.

What cadAI is made of

cadAI is not one standalone program. It is three forms plus one assistant, all running on the same kernel. The web version opens at cad.aiglade.cn and works in the browser, with nothing to install. Its core work is primitives, holes, conversion, measurement, and preview. The desktop version is local modeling with the same Rust kernel, for people who need local files. The MCP interface is the third form, and it is the subject of this page.

The difference between the MCP interface and the assistant is who is speaking. With the assistant, a person speaks one sentence in the web version. With MCP, an agent calls intents from a program. The agent does not click buttons on a web page. It breaks one sentence into steps, calls the intents in order, and builds the model directly on the same kernel.

What an agent can do, and what it cannot do

An agent can do exactly what the five intents can do, no more and no less. S-01 primitives builds the rough shape: a flanged plate starts as a disc. S-02 holes drills where holes belong, with the count and position following the parameters. S-03 conversion handles export, with the format list to be checked on the page. S-04 measurement checks the key dimensions, and S-05 preview shows the model before delivery. One sentence becomes several steps: build a flanged plate with a ring of holes, and the agent builds the shape, drills the holes, measures, and previews, in that order.

Parameters are where an agent and a person most often go wrong. The hole count, the hole positions, and the thickness must be stated clearly, because the agent works from the parameters it is given and does not invent a number to fill a gap. If a parameter is unclear, the agent should ask it back rather than guess. An agent cannot do what cadAI itself cannot do. Complex curved surfaces, assembly motion simulation, and CAM toolpaths are outside the scope, so an agent cannot do them either. Chamfer is not among the five connected intents, so this page does not write it as supported.

Three kinds of choices

There are three kinds of choices, and the table shows them at a glance. This page compares categories only: it gives no version numbers, prices, or scores for any specific product.

ChoiceInstallParametric modelingHoles and measurementAI agent driveOne sentence modelingWho it fits
cadAI web version and MCPOpens in the browserYes, the part follows the parametersS-02 holes, S-04 measurementMCP interface, an agent can drive modelingAI assistant, calls S-01 to S-05Simple parts, flanged plates, brackets, agent built models
Desktop CADInstall local softwareYesYesUsually not offeredUsually not offeredComplex curved surfaces, assemblies, CAM
General online toolsOpens in the browserMostly simple editsDepends on the toolUsually not offeredUsually not offeredLight viewing and small edits

Choose by the work, not by the novelty. A simple part that an agent should build for you is enough for cadAI, and a person without coding experience can use the one sentence assistant instead of MCP. A part with complex curved surfaces, or work needing assembly motion simulation or CAM toolpaths, belongs in desktop CAD. Opening a model to look at it once is what general online tools are for.

Let an agent build a flanged plate in 6 steps

Take a flanged plate as the example, and walk through 6 steps:

  1. Connect the agent to the MCP interface. The developer sets up the agent side from the interface notes, and the setup details follow the official notes.
  2. Give the agent one sentence: build a flanged plate with a ring of holes. State the outline, the thickness, the hole count, and the hole positions clearly, and do not let the agent guess.
  3. The agent calls S-01 primitives to build the rough shape. A flanged plate starts as a disc, and the rough shape comes before detail.
  4. The agent calls S-02 holes to drill where holes belong. Their count and position follow the parameters, so a changed parameter moves them with it.
  5. The agent calls S-04 measurement and S-05 preview. Measure the key dimensions, look at the preview, and fix the parameters before export if the dimensions are wrong.
  6. A human previews and confirms the draft, and only then exports it. The model must not go straight to production, and the hand check of dimensions and tolerances comes first.

Step 6, the human confirmation, is the one most often skipped by mistake. However smoothly the agent works, a person still needs to look before export. A parameter change moves holes, and a moved hole changes dimensions, so export without measuring again turns a wrong dimension into a wrong part. MCP saves the work of clicking buttons, not the human look that confirms the model is right.

cadAI in numbers

Evidence

cadAI is a browser-based parametric CAD from Aiglade, for mechanical part modeling, hole drilling, measurement, and export, with the same kernel on desktop, in the browser, and over MCP. Its kernel is SolidCraft, an independent community project, not an ArtCraft product line. Its AI assistant takes 5 intents: S-01 primitives, S-02 holes, S-03 conversion, S-04 measurement, and S-05 preview. The MCP interface is built for AI agents, and an agent drives modeling directly through it, calling the same 5 intents. The result is a draft, and a human previews and confirms it before export, because pricing is still to be confirmed and no price, tier, or free allowance is stated anywhere in this page.

FAQ

Q: Can cadAI let an AI agent build models through MCP?

Yes. The MCP interface is built for AI agents: an agent drives modeling directly through it, calling the five connected intents, S-01 primitives, S-02 holes, S-03 conversion, S-04 measurement, and S-05 preview. The result is a draft, and a human previews and confirms it before export.

Q: Can the agent save the model straight to disk?

No. The agent produces a draft, and delivery is preview, confirm, then export. A parameter change can move a hole without anyone noticing, so export without a human look turns a wrong dimension into a wrong part.

Q: I have no coding experience. Can AI still build a model for me?

Yes, use the one sentence assistant in the web version, not MCP. Open cad.aiglade.cn and tell the assistant one sentence, such as building a flanged plate disc, drilling a ring of holes, then measuring the spacing. MCP is the developer channel, and the assistant is for people who do not write code.

Not written / to verify

cadAI · Aiglade · 2026-11-12

Open cadAI