MCP Tools Overview

The vcad MCP server exposes the parametric kernel as a set of tools that AI agents can call through the Model Context Protocol. The server runs locally via npx @vcad/mcp and includes the full WASM engine, so geometry evaluation happens on your machine.

This page lists every available tool with its purpose and key parameters. For setup instructions, see MCP Server Setup. For detailed usage of the core design tools, see Creating Models with AI.

Design Tools

create_cad_document

The primary tool for building 3D geometry. Takes a structured JSON description of parts and operations, and returns an IR document that can be inspected, exported, or opened in the browser.

Each part can be defined by a primitive (cube, cylinder, sphere, cone), or by a sketch-based operation (extrude, revolve, sweep, loft). Parts support a chain of operations that modify the geometry: boolean operations (union, difference, intersection), transforms (translate, rotate, scale), patterns (linear, circular), features (fillet, chamfer, shell), and holes.

The tool also accepts an optional assembly block with instances and joints for multi-body models.

Positions throughout the tool accept three formats: absolute coordinates ({x: 25, y: 15, z: 0}), named positions ("center", "top-center", "bottom-center"), or percentage values ({x: "50%", y: "50%"}).

For complete documentation, see create_cad_document reference.

create_cad_loon

An alternative to create_cad_document that accepts geometry as Loon source code -- a compact Lisp-like language designed for parametric CAD. Loon is often more concise than the JSON format, especially for models with many chained operations.

[let plate [cube 50 30 5]]
[let hole [translate 25 15 0 [cylinder 2.75 10]]]
[let result [difference hole plate]]
[root result "aluminum"]

Loon supports let bindings, pipe for operation chaining, all primitives, booleans, transforms, and features. See Loon Format and the create_cad_loon reference for syntax details.

open_in_browser

Generates a shareable URL that opens an IR document in the vcad.io web app. The document is compressed (gzip + base64url) and embedded in the URL, so no server storage is needed.

Returns a URL like https://vcad.io/#doc=.... Note that very large documents may exceed browser URL length limits (roughly 2KB of compressed data).

Analysis Tools

inspect_cad

Evaluates an IR document through the kernel and returns geometric properties. This is the main tool for validating that a design meets dimensional and physical requirements.

Returns:

The AI typically calls inspect_cad immediately after create_cad_document to verify the geometry before exporting. If something looks wrong (unexpected volume, bounding box doesn't match intended dimensions), it can adjust the parameters and try again.

For detailed documentation, see inspect_cad reference.

export_cad

Exports an IR document to a file on disk. Format is determined by the filename extension.

See export_cad reference for details.

STEP Tools

import_step

Reads a STEP file (AP203 or AP214) and returns an IR document with ImportedMesh nodes. The IR can then be inspected, exported to other formats, or combined with new geometry.

Supports files from common CAD tools: Fusion 360, SolidWorks, Onshape, CATIA, Creo, and others that produce AP203/AP214 STEP.

See STEP I/O reference for details.

Simulation Tools

The simulation tools provide a gym-style reinforcement learning interface built on the Rapier3D physics engine. You create an environment from a vcad assembly (which defines rigid bodies via parts and kinematic constraints via joints), then step the simulation with actions and observe the resulting state.

create_robot_env

Creates a physics simulation environment from a vcad assembly document. The assembly must contain part instances and joints. Returns an env_id string that identifies the environment for subsequent calls.

gym_step

Advances the simulation by one timestep with the provided action.

Returns an observation object with joint positions, joint velocities, end effector poses, a reward value, and a done flag.

gym_reset

Resets the simulation to its initial state. Returns the initial observation.

gym_observe

Returns the current observation without advancing the simulation. Useful for reading state between steps or after a reset.

gym_close

Destroys a simulation environment and frees its resources.

Batch Operations

For parallel reinforcement learning, three batch tools create and control multiple environments simultaneously:

batch_create_envs creates N copies of the same environment. Returns a batch_id.

batch_step steps all environments with per-environment actions. Returns observations, rewards, and done flags for every environment.

batch_reset resets all environments in the batch to initial state.

Electronics Tools

The MCP server also includes tools for electronic design (schematic capture, PCB layout, and fabrication export). These operate on a separate data model embedded in the vcad document.

Utility Tools

get_changelog

Queries the vcad changelog for recent changes, filtered by version, category, feature tag, or MCP tool name. Useful for discovering new capabilities.

Error Handling

When a tool call fails, the server returns an error response that the AI can use to self-correct:

{
  "content": [{"type": "text", "text": "Error: Invalid primitive type 'box', use 'cube' instead"}],
  "isError": true
}

Common errors include invalid primitive types, missing required fields, and file I/O failures. The error messages are designed to be actionable -- they tell the AI what went wrong and often suggest the fix.

For setup instructions, see MCP Server Setup. For a tutorial on the design workflow, see Creating Models with AI.