Meshy 3D Agent

AI agent skills for the Meshy AI 3D generation platform. Enables AI coding assistants (Cursor, Claude Code, OpenClaw) to generate 3D models, textures, images, rig characters, animate them, and prepare models for 3D printing — no MCP server required.

New Updates: For new web agent experience, come to try out our new Meshy Workspace Agent! It supports:

• 🧠 Brainstorm with you, pitching directions first
• 🎨 Generate visuals in batches, refined through chat
• 🧊 Turn your favorite into 3D — print, download, anywhere
• 📚 Answer your 3D questions in-line

How It Works

These are agent skills — no server, no MCP process. Each skill bundles a small Python CLI (scripts/) plus on-demand markdown references (references/), and your AI assistant reads the skill files and drives the Meshy API directly through the bundled scripts.

Skills

meshy-3d-generation (core)

Full 3D generation lifecycle: API key setup, task creation, polling, downloading, and multi-step pipelines.

Capability	Description	Credits
Text to 3D	Generate 3D models from text descriptions	20-30
Image to 3D	Convert single or multiple images to 3D	20-30
Retexture	Apply new textures to existing models	10
Remesh	Change topology, polycount, or export format	5
Convert	Convert a model to other formats (glb/fbx/obj/usdz/blend/stl/3mf) without remeshing	1
Resize	Rescale to a real-world size (height / longest-side / auto)	1
UV Unwrap	Generate a clean UV layout for external texturing (GLB, ≤40k faces)	5
Auto-Rigging	Add skeleton to textured humanoid characters, ≤300k faces (includes walking + running)	5
Animation	Apply custom animations to rigged characters (action_id from the public Animation Library)	3
Text to Image	Generate 2D images from text (recommended pre-step before image-to-3d)	3-9
Image to Image	Optimize/edit reference images (recommended pre-step before image-to-3d)	3-12

meshy-3d-printing (optional)

3D printing workflow: slicer detection, automated printability analysis & repair, white model printing, multicolor printing via API.

Capability	Description	Credits
White Model Print	Generate → OBJ download → coordinate fix → slicer launch	20
Multicolor Print	Generate → texture → multi-color API → 3MF → slicer launch	40
Creative Lab	One tool: photo or text → finished printable product (figure / lamp / keychain / fridge-magnet); prototype→build end-to-end	36
Slicer Detection	Auto-detect 7 slicers: OrcaSlicer, Bambu Studio, Creality Print, Elegoo Slicer, Anycubic Slicer Next, PrusaSlicer, UltiMaker Cura	0
Analyze Printability	Automated FDM check via /openapi/v1/print/analyze (watertight / volume / non-manifold / degenerate / holes)	0 (free)
Repair Printability	Fix non-manifold edges, degenerate faces, holes via /openapi/v1/print/repair (output format mirrors input)	10

The printing skill bundles the same task-runner CLI as the generation skill and shares its environment setup.

meshy-openclaw (OpenClaw / ClawHub)

A single unified skill for the OpenClaw ecosystem. Combines generation + printing into one self-contained skill, with OpenClaw-compatible metadata.clawdbot frontmatter and a full SECURITY MANIFEST.

Capability	Description	Credits
All generation	Text/Image to 3D, Creative Lab, Retexture, Remesh, Convert, Resize, UV Unwrap, Rig, Animate, Text/Image to Image	1–36
3D printing	White model (OBJ) + multicolor (3MF via API) + slicer detection	0–35

Designed for ClawHub publishing. API key is stored only in .env in the current working directory — no shell profile access.

Quick Install

One command to install all skills:

npx skills add meshy-dev/meshy-3d-agent

Then set your API key (pick any method below):

Or simply start using the skill — when the agent loads it, it will detect that no API key is configured, ask you for it, and set it up for the current session (it never writes your key anywhere except .env on your request).

macOS / Linux

Option A: Global (recommended) — add to your shell profile so it persists across sessions:

nano ~/.zshrc

Add this line at the end, save and exit (Ctrl+O, Enter, Ctrl+X):

export MESHY_API_KEY="msy_YOUR_API_KEY"

Then reload:

source ~/.zshrc

Option B: Project-local — create a .env file in your project root:

echo 'MESHY_API_KEY=msy_YOUR_API_KEY' > .env

Remember to add .env to your .gitignore to avoid committing your key.

Windows

Option A: Permanent (recommended) — set via System Environment Variables:

1. Open Settings → search for "Edit environment variables for your account"
2. Add a new user variable named MESHY_API_KEY with your key as the value
3. Restart your terminal for it to take effect

Option B: Current session only:

$env:MESHY_API_KEY = "msy_YOUR_API_KEY"

Option C: Project-local — create a .env file in your project root:

MESHY_API_KEY=msy_YOUR_API_KEY

Prerequisites

• A Meshy API key (get one here — requires Pro plan or above)
• Python 3 with requests package (pip install requests)

Manual Installation

OpenClaw

Option A: Via ClawHub

npx clawhub install meshy-dev/meshy-3d-agent

Option B: Manual — copy the skills/meshy-openclaw/ folder to your OpenClaw skills directory.

Cursor

mkdir -p .cursor/skills

# Core (required)
cp -R skills/meshy-3d-generation .cursor/skills/

# 3D Printing (optional)
cp -R skills/meshy-3d-printing .cursor/skills/

Claude Code

mkdir -p .claude/skills

# Core (required)
cp -R skills/meshy-3d-generation .claude/skills/

# 3D Printing (optional)
cp -R skills/meshy-3d-printing .claude/skills/

Codex

Codex reads skills from .agents/skills — per repository, or from ~/.agents/skills to make them available everywhere.

mkdir -p .agents/skills

# Core (required)
cp -R skills/meshy-3d-generation .agents/skills/

# 3D Printing (optional)
cp -R skills/meshy-3d-printing .agents/skills/

.agents/skills is the cross-editor convention, so Cursor picks these up as well.

Skill vs MCP Server

Feature	Agent Skill (this repo)	MCP Server
Setup	Copy a skill directory	npx meshy-mcp-server
Dependencies	Python 3 + requests	Node.js >= 18
How it works	AI reads instructions, makes API calls directly	Dedicated server process with structured tools
IDE support	Amp, Cline, Codex, Cursor, Gemini CLI, Claude Code, OpenCode and 20+ more	Any MCP-compatible client
File management	Via skill instructions	Built-in auto-save with project folders

Both approaches provide the same Meshy API capabilities. Choose based on your preference and setup.

Security & Data

These skills run entirely on your machine and talk to a single service — the Meshy API. No telemetry, no third-party endpoints.

Your API key

• Read from the current session environment, or from .env / .env.local in the current working directory. Home directories and shell profiles are never scanned.
• Sent only in the HTTP Authorization: Bearer header to https://api.meshy.ai. It is never logged in full — scripts print at most a msy_1234... prefix.
• Never persisted by the scripts. The key is written to .env in the current working directory only when you explicitly ask, and that .env is added to .gitignore automatically. It is never written to shell profiles, Windows user variables, or any path outside the working directory. Persisting it globally is offered to you as instructions to run yourself — the skill never does it silently.
• System proxies are bypassed (requests session trust_env = False), so the key is never handed to an environment-configured proxy.

What leaves your machine

• Only what a generation request needs: your API key, text prompts, and image URLs/data (for image-to-3D) — all to api.meshy.ai.
• Generated assets are downloaded and saved locally under ./meshy_output/; nothing else is uploaded.

Filesystem footprint

• Reads: .env / .env.local in the working directory, and any input files you explicitly pass (e.g. local images), at the exact path you provide.
• Writes: ./meshy_output/ (models, thumbnails, metadata.json, history.json) and — on request — .env in the working directory.
• The 3D-printing skill additionally launches an already-installed slicer with your model file; it never downloads or installs software.

For Maintainers

Single sources of truth — edit these, never the generated copies:

Source	Generated outputs
reference/source.md	skills/*/reference.md (the OpenClaw build also injects the SECURITY MANIFEST extracted from skills/meshy-openclaw/SKILL.md)
scripts/src/meshy_task.py	skills/*/scripts/meshy_task.py
skills/meshy-3d-printing/scripts/slicers.py, fix_obj.py	skills/meshy-openclaw/scripts/slicers.py, fix_obj.py

After editing a source, regenerate and verify:

python3 scripts/build.py          # regenerate all targets
python3 scripts/build.py --check  # CI mode: fail if outputs are stale, a SKILL.md exceeds 300 lines, or a references/*.md is unlinked

Generated files carry a GENERATED marker comment. CI runs python3 scripts/build.py --check to reject edits that bypass the sources.

License

MIT