agentcad logo

agentcad

Freemium
訪問

An open-source local CAD CLI and MCP server for AI agents to build and inspect CAD models.

agentcadとは?

agentcad is an open-source CAD command-line tool and MCP server that lets coding agents create and inspect models through Python scripts. It runs build123d by default and supports CadQuery as an optional runtime. Its central workflow is iterative: an agent runs a model script, receives geometry metrics and rendered previews, then uses that feedback to refine the design. Projects can retain numbered versions and compare later iterations, with generated files kept locally.

agentcadでできること

Scripted CAD execution

Run Python CAD scripts with build123d primitives, `show_object()`, and agentcad edit helpers pre-injected, so scripts do not need setup imports.

Geometry checks and metrics

Static checks identify syntax errors, missing `show_object()`, and unresolved imports before the CAD engine runs. Successful builds report dimensions, volume, surface area, face and edge counts, and validity.

Rendered review

Successful runs can produce a four-view PNG composite and a review viewer, giving an agent visual output to inspect alongside numeric geometry measurements.

Version history and diffs

Each successful run is saved in a numbered version directory rather than overwriting prior output. From the second version onward, agentcad can compare iterations with side-by-side, overlay, and volume-difference views.

CAD export and inspection

The tool exports STEP as part of its core workflow and supports STL, GLB, and OBJ outputs. `measure` reads feature sizes from finished files, while `inspect` provides topology information.

Agent setup and integration

The MCP server can be configured for clients including Claude Code, Cursor, and Windsurf. `agentcad init` adds project guidance as an AGENTS.md block or CLAUDE.md content and installs a Claude Code skill.

利用シーン

“Create printable parts from prompts”

Use a coding agent to turn dimensions and requirements for objects such as stands, enclosures, or bedside docks into a model, then inspect the preview and export an STL for printing.

“Iterate on a CAD design”

Keep successive model runs as separate versions and use geometry metrics and visual diffs to see how a change affected the part.

“Review model validity before fabrication”

Check reported dimensions and validity, and use renders or topology inspection to catch geometry problems before handing a design to a printer or another CAD workflow.

“Build models for web or device viewing”

Export GLB or OBJ for web viewers, or request USDZ through an agent workflow such as the documented Muse example for viewing a model on an iPhone.

よくある質問

Which Python versions are supported?

agentcad supports Python 3.10 through 3.12. Its OpenCascade bindings are not available for Python 3.13, according to the documentation.

Can I use it without an MCP client?

Yes. Install the CLI with `pip install agentcad`. To use the MCP server, install the `mcp` extra with `pip install 'agentcad[mcp]'` or configure a client to launch it through `uvx`.

What does a successful run produce?

A successful run registers a numbered version and produces a STEP file with run metadata and geometry metrics. Preview, viewer, and other optional artifacts report their own status separately; an optional artifact failure does not undo a successful core build.

Can it run CadQuery scripts?

CadQuery is supported as an optional extra with separate dependencies. The default installation uses build123d.

クイック情報

Product type
Open-source CAD CLI and MCP server
Primary runtime
build123d; CadQuery is optional
Supported Python
3.10–3.12
Runs
Locally
License
Apache-2.0
Project domain
agentcad.dev

agentcad のトラフィック分析

トラフィックデータは参考情報としてご利用ください。

月間訪問数
0
世界ランキング
-
直帰率
0.00%
平均滞在時間
00:00
訪問あたりのページ数
0.00
ドメイン評価
10

トラフィック推移

トラフィックソース

トラフィック分析データはまだありません。

主な地域

トラフィック分析データはまだありません。