Meshy Plugin

Name: meshy

Description: Generate 3D models, textures, rigs and animations from text or images via the Meshy AI API — a zero-dependency Node client wrapped in a skill

Author: Flop (flopspm@gmail.com)

Version: 0.1.2

Keywords: meshy, 3d, ai, text-to-3d, image-to-3d, generative, model-generation

The Meshy plugin gives Claude Code access to the full Meshy AI 3D generation API through a single zero-dependency Node script (meshy.mjs) wrapped in a model-invoked skill. It deliberately avoids the official MCP server: an MCP server loads all of its tool schemas into context on every session, while a skill loads only its one-line description until it's actually needed — far cheaper for an API you reach for occasionally. The script runs identically on macOS, Linux, and Windows (it only needs Node), with no npm install and no build step.

Installation

claude plugin install meshy@flugins

Important: After installing the plugin, restart Claude Code to activate skills.

Tip: Enable auto-update via /plugin → Installed → select the plugin → enable auto-update.

Requirements

• Node.js 18+ on PATH (node --version) — the script uses native fetch and streams; no other runtime or dependency is required.
• A Meshy API key in the MESHY_AI_API_KEY environment variable (or MESHY_API_KEY, or passed via --api-key). Create one at meshy.ai/settings/api. Keys look like msy_....
• Credits on your Meshy account for paid operations. A free, no-cost test key (msy_dummy_api_key_for_test_mode_12345678) returns sample assets without spending credits — handy for dry-runs.

Features

Skills

• Meshy 3D Generation - Generate 3D models, textures, rigs, animations and images from text or photos, check credit balance, and download finished assets

Usage

Ask in natural language — "make a 3D model of a red ceramic mug", "turn this photo into a 3D model", "rig this character", "how many Meshy credits do I have?" — and the skill activates automatically. It picks the right endpoint, runs the bundled CLI (which handles the asynchronous create → poll → download lifecycle internally), parses the single JSON object the script prints, and reports where the files were saved and how many credits were spent. For long jobs it can fire-and-forget with --no-wait and resume later with status / download.

Configuration

No plugin-specific configuration. The skill reads the API key from the environment (MESHY_AI_API_KEY / MESHY_API_KEY) or a --api-key flag, and the script ships sensible defaults (20-minute poll timeout, 5-second poll interval, automatic retry with backoff on rate limits). The full endpoint and parameter reference lives in skills/meshy/docs/api-reference.md and is loaded on demand rather than into every session.

---

Meshy 3D Generation

Skill: meshyType: Model-invoked (automatic) / user-invocable

Drives the entire Meshy REST API from one generic client. The script exposes every endpoint behind a uniform create → poll → download engine and prints exactly one machine-readable JSON object to STDOUT (progress goes to STDERR), so Claude never has to scrape human text or write its own polling loop.

How it Activates

The skill activates whenever you mention 3D generation or Meshy. Examples:

• "generate a 3D model of a low-poly tree"
• "turn this image into a 3D mesh"
• "re-texture this model to look like weathered bronze"
• "rig and animate this character"
• "check my Meshy balance"

You can also invoke it explicitly from the / menu.

Capabilities

Area	Commands
3D generation	text-to-3d (+ refine), image-to-3d, multi-image-to-3d
Post-processing	retexture, remesh, rig, animate, convert, resize, uv-unwrap
Image generation	text-to-image, image-to-image
3D printing	analyze-printability (free), repair-printability, multi-color-print
Experimental	creative-lab (keychain / fridge-magnet / figure / lamp, prototype → build)
Account & tasks	balance, status <type> <id>, list <type>, download <type> <id>

Key Behaviors

• Asynchronous handled for you. Every generation POST returns a task id; the script polls until the task reaches SUCCEEDED/FAILED/CANCELED, streaming progress to STDERR. Use --no-wait to return the id immediately and resume later.
• Downloads with --out. Result URLs are short-lived signed links (models are retained for only a few days). Pass --out <dir> to stream assets to disk immediately; --formats glb,fbx selects model formats (default glb), and --textures / --thumbnail / --all pull extras. If a URL has expired, download re-fetches the task once for fresh links.
• Local images, no manual base64. Pass a local file to --image / --images / --texture-image-url and the CLI encodes it into a data URI for you.
• Whole API reachable. Common parameters have typed flags; anything exotic is reachable via --param key=value or --json '{...}'.
• Clear failure signals. Exit codes distinguish auth (3), insufficient credits (4), rate-limit (5), task-failed (7), timeout (8), and network (9), and the JSON error.kind mirrors them.

Credits (approximate)

Preview text-to-3D ≈ 5 (meshy-5) to 20 (meshy-6 / low-poly); refine ≈ 10; image-to-3D ≈ 20–30; remesh / rig ≈ 5; animate ≈ 3; convert / resize ≈ 1; analyze-printability is free. Run balance before a large job; an exit code 4 means you're out of credits.

What the Skill Will NOT Do

• Never spends credits on a non-trivial job without confirming with you first
• Never writes its own polling loop — the script owns the lifecycle
• Never hand-rolls base64 for images — local paths are encoded by the CLI
• Never requires an MCP server or any npm dependency

See the SKILL.md reference and the bundled docs/api-reference.md for the full command and parameter specification.