MCP App Store
by KeenEthics
Arpelier icon

Arpelier MCP Server for Claude

By Arpelier

Overview

Arpelier is a process-flow workspace for semiconductor and nano devices, made for people and AI to work on together. Each step of a process flow (deposition, lithography, etch, implant, CMP and more) becomes a 3D structure you can inspect. This connector is the AI side of that workspace: Claude writes a process flow as Flow 1.0 JSON and checks it with the same engine the Arpelier web app uses, so it can measure the result instead of estimating it. With the connector, Claude can: - read Arpelier's docs for writing flows: process kinds and fields, material IDs, mask polarity and examples; - validate a flow with the web app's own parser, with each problem located in the JSON; - compute the structure up to any step and report material volumes and the surface height range; - take cross-sections with exact layer stacks (material, bottom and top in nm); - check whether two materials touch and their shared area, or whether points are connected; - list the material information that a 3D export (GLB, 3MF, STL) would carry; - make a link that opens a checked flow, with its computed result, as a new unsaved project in the web app. Six of the seven tools are read-only. The connector needs no account or sign-in and never opens or changes your Arpelier projects. It keeps data in two cases only: the computed structure (no names or notes) for up to one hour, to continue long or module-by-module calculations; and, for a link you ask for, the project made from the flow (with names and notes) and its computed result for 7 days. Anyone with a link can open it until it expires. Projects are saved only in the web app (arpelier.com). Results are geometric, with lengths in nm: Arpelier is for understanding, teaching and exploring process-structure relationships, not a fabrication recipe, a TCAD simulation or an electrical sign-off. Each calculation can use about 160,000 lateral grid columns and 400 executed steps; one longer than about 3 minutes continues over several calls.

How to connect

  1. Claude: Open the Arpelier listing in the Claude connectors directory and choose Connect. Claude listing Arpelier exposes 7 tools here.

Tools

arpelier_compute_flow

Claude
Really compute the Flow with the Arpelier engine up to stepId (default final). Returns the executed step count, grid, per-material volume and fraction, top-surface height range (min/max/mean, nm) and voids: the number of separate regions of empty space below the surface (cavities, or gaps under an overhang such as a released sheet), their total volume and the bounds of the largest; restoredSteps counts the steps continued from a computed state kept from an earlier call. A calculation longer than one call returns IN_PROGRESS, and the same call again continues it. Read-only: writes to no project.

arpelier_cross_section

Claude
Compute the Flow up to stepId and sample a section in project coordinates (nm). axis "x" cuts at x = at (shows y–z), "y" cuts at y = at (shows x–z), "z" is a plan view at height at. Returns a legend, a character map (top row = highest z) and, for vertical sections, exact layer stacks read from the engine (material, fromNm, toNm, thicknessNm) for each distinct column type with the horizontal rangesNm where it occurs. For vertical sections the map and stacks are drawn from exact engine boundaries (every layer appears; layers thinner than a character row are shown as one row), so dz only affects the raw includeMatrix sampling. The most common column types are listed first; when the stacks would be long (tall layer stacks, many column types) the rest are counted in moreStacks. includeMatrix adds raw material IDs for small sections.

arpelier_export_structure

Claude
Compute the Flow up to stepId and return the material info that a 3D structure export carries (materials.json, schema arpelier.export-materials/1.0): per material its type, built-in doping family, grid-cell volume, height range, bounds in nm and the steps that deposit it. Returns no file: GLB, 3MF and STL files are made in the browser with window.arpelier.exportStructure.

arpelier_inspect_structure

Claude
Compute the Flow up to stepId and run read-only structure checks with the same request contract as window.arpelier.inspectStructure (see /docs/agents/tasks/inspection.md). Every check has a unique id and a type. "interface" takes materialA, materialB, boundsNm {x:[min,max], y:[min,max], z:[min,max]} and minAreaNm2 (maxAreaNm2 optional) and returns the shared-face area in nm². "connection" takes materialIds, connected (the expected answer) and 2–8 anchors {xNm, yNm, zNm, materialId}. Material IDs are exact, as computed: a substrate given material 5 with substrateDoping "p" is Si (p), ID 4 (read the legend of arpelier_cross_section). Up to 16 checks.

arpelier_open_in_app

Claude

arpelier_read_docs

Claude
Return Arpelier's published reference documentation (Markdown or JSON): the Flow format, process kinds and fields, material IDs, mask polarity, example Flows and the web app's browser API. With no path it returns the index (/llms.txt). Paths are the links used in the docs, e.g. "/docs/agents/processes.md", "masks", "tasks/repetition", "examples/oxide-window.flow.json", "schema/flow-1.0.json". Long pages continue with offset.

arpelier_validate_flow

Claude
Validate a Flow 1.0 document with the web app's own parser (same report as window.arpelier.validate for an input document). Returns valid plus issues with JSON paths. Does not compute.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/arpelier"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/arpelier)