Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@mcp-cad-viewer Show me bracket-block.stl in the 3D viewer"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
An MCP server that opens CAD parts in an interactive 3D viewer inside the chat, using MCP Apps.
Most MCP tools answer with text. For mechanical engineers that's the wrong shape: "the bracket is 20 × 10 × 5 mm with 12k triangles" is less useful than the bracket itself. view_part returns a summary the model can reason about and a View the host renders in a sandboxed iframe, where you can rotate and zoom the part.
Status: early. The server, the STL pipeline and the View are working and smoke-tested. Next up is testing in real MCP Apps hosts and adding STEP support. See the roadmap.
model ──calls──▶ view_part(path) host (Claude, VS Code, ChatGPT…)
│ │
├─ text: "12 triangles, 20 × 10 × 5" ───▶│ shown to the model
├─ structuredContent: summary + STL ───▶│ passed to the View
└─ _meta.ui.resourceUri ──────────────────▶ fetches ui://cad-viewer/viewer.html
└─ renders it in a sandboxed iframe
(three.js, orbit controls)Server (src/server.ts): MCP TypeScript SDK v2. serveStdio negotiates the protocol era per connection, so the same server works with 2025-era clients and stateless 2026-07-28 clients.
View (view/): uses the App class from @modelcontextprotocol/ext-apps to receive the tool result and three.js to render it. It's bundled with esbuild and inlined into one HTML resource, so it needs no network access and no CSP exceptions.
Safety: files are only read from inside CAD_VIEWER_ROOT. Path traversal is rejected, and parts are capped at 8 MB.
Related MCP server: ifc-mcp
npm install
npm run smoke # builds, starts the server, calls view_part on samples/bracket-block.stlAdd it to an MCP Apps–capable host (stdio):
{
"mcpServers": {
"cad-viewer": {
"command": "npx",
"args": ["-y", "mcp-cad-viewer"],
"env": { "CAD_VIEWER_ROOT": "/path/to/your/parts" }
}
}
}Then ask: "Show me bracket-block.stl".
view_part tool with STL parsing (ASCII and binary), bounding box and triangle count
Bundled three.js View delivered as an MCP Apps ui:// resource
Smoke test through a real MCP client
Release workflow: tag to publish on npm (with provenance) and the official MCP Registry
Verify rendering in Claude, VS Code and ChatGPT; follow the host's theme
STEP and IGES support (OpenCascade.js, converted server-side to a mesh)
Measurement tools in the View: click two points for a distance, section planes
View-to-server calls: "highlight the face I clicked" goes back to the model as context
Streamable HTTP transport and a hosted demo on free infrastructure
Shrink the View bundle (tree-shake three.js, lazy-load loaders)
fusion-mcp-server: drives Autodesk Fusion from an agent; this viewer shows the result
mcp-tool-bench: conformance checks and tool-selection evals for MCP servers
MIT © Ashutosh Singh
Open an STL part in an interactive 3D viewer (rotate, zoom, measure the bounding box). Returns a text summary with triangle count and dimensions as well.
| Name | Required | Description | Default |
|---|---|---|---|
| path | Yes | Path to an .stl file, relative to the viewer's root folder |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are supplied, so the description carries the full behavioral burden. It does disclose the side effect of opening an interactive viewer and the fact that a text summary is also returned, which is genuinely useful. However, it says nothing about whether the viewer blocks, whether a GUI/display is required, file-size limits, or how missing or malformed .stl files are reported.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two compact sentences with the primary action front-loaded and the return behavior appended where it belongs. Nothing is padded, though the parenthetical capability list is slightly dense and could be ordered more usefully.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a one-parameter, no-output-schema tool, the definition covers the action, the interactive capabilities, and the shape of the returned summary. The only meaningful omission is error and environment behavior, which matters but is secondary for a read-only viewer.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100% and the single 'path' parameter is already documented as root-relative in the schema, so the baseline of 3 applies. The description adds only weak reinforcement ('STL part') and no extra semantics such as accepted extensions or path normalization rules.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description names a specific verb and resource ('Open an STL part in an interactive 3D viewer') and states what the call yields (a text summary with triangle count and dimensions). There are no sibling tools to disambiguate against, so the only thing keeping it from a 5 is that the phrase 'view_part' vs. the interactive viewer's scope is not further bounded.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Usage is only implied by the description of the action; there is no explicit statement of when to choose this tool, what prerequisites exist (a viewer session, a valid root-relative file), or what to do on failure. With no siblings, the absence of alternatives is acceptable, but the lack of preconditions leaves guidance thin.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0view_partScored across 1 tool
With only one tool, there is no possibility of misselection; its purpose (open an STL in an interactive 3D viewer) is unambiguous and self-contained.
The single name view_part follows a clear verb_noun snake_case convention that reads predictably and would scale well if more tools were added.
One tool is thin for a CAD viewer surface; realistic use cases like re-viewing, closing a session, or exporting a screenshot have no counterpart, though the scope itself is narrow enough that a single entry point is defensible.
The tool covers a single format (STL) and a single action (view/measure), with no support for other CAD formats, section/measurement beyond the bounding box, or capture/export of the view. These are notable gaps an agent cannot work around, though the core stated purpose is met.
CAD in ChatGPT and Claude: editable parts, STEP/STL/drawings, in-chat 3D, automatic checks.
Parametric 3D CAD for AI agents: build print-ready parts, check them, export STL, 3MF or STEP.
Fast multi-format CAD previews, geometry checks and conversion; measure and share in the workspace.
DXF and PDF/X-4 for AI agents: structured facts, PNG renders, an interactive in-chat viewer.
