_{GPT-5 xhigh · authentic 14-stage build · 58 bones · 2,453 generated cubes · 11,524 exported triangles · 1 Meshopt GLB primitive}

ashfox — AI-native low-poly asset workbench

Describe an asset and watch your AI agent model, texture, rig, and animate it in an editable browser workbench. Keep the canonical .ashfox source, then export runtime-optimized Bedrock, GeckoLib 5, glTF, or GLB output. A fully supported Blockbench MCP workflow is included for teams already using Blockbench.

Explore ashfox.io →  ·  Open the demo gallery →
_{Build complete low-poly assets with your AI agent—no modeling app required.}

Connect Blockbench  ·  Use without Blockbench  ·  Demo gallery  ·  Read the guides

Choose the workflow you want

Both workflows are open source and local-first. The web workbench is optional; you do not need to switch away from Blockbench.

Connect Blockbench

1. Install the plugin

In the latest Blockbench Desktop:

1. Open File → Plugins → Load Plugin from URL.
2. Paste the release URL below.
3. Choose Install or Load.

https://github.com/sigee-min/ashfox/releases/latest/download/ashfox.js

This URL loads the published ashfox.js asset from this GitHub repository. The plugin opens a local MCP endpoint on your machine; it does not require an ashfox account or a connection to ashfox.io.

2. Connect your agent

Keep Blockbench open, then add this HTTP MCP endpoint to your agent or MCP client:

http://127.0.0.1:8787/mcp

Ask the agent to confirm the connection before it edits anything:

Connect to the Blockbench MCP endpoint, call list_capabilities, inspect the
current project, and tell me which modeling, texture, animation, preview, and
export tools are available. When the connection check is complete, ask me
exactly: "What would you like to create?" Treat my next message as the complete
asset brief and begin unless one required target detail is missing. Do not
change the project until I answer.

3. Make your first asset

Once the agent reports that Blockbench is connected:

Create a Minecraft-style fantasy creature with a readable silhouette,
consistent pixel textures, a clean bone hierarchy, and an idle animation.
Validate it and render a preview before exporting.

That is the complete setup. You can stay entirely inside Blockbench.

curl -s http://127.0.0.1:8787/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

curl -s http://127.0.0.1:8787/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"list_capabilities","arguments":{}}}'

What the Blockbench integration exposes

• Modeling: bones, cubes, meshes, transforms, updates, and deletes.
• Texturing: assignment, deterministic face painting, mesh painting, and texture reads.
• Animation: clips, frame poses, triggers, and keyframes.
• Review: project state, validation, capability inspection, and rendered previews.
• Delivery: export through the active Blockbench project and format.

Normal edits use revisions so an agent cannot silently apply work against stale project state. Use only trusted local MCP clients because connected tools can edit the project currently open in Blockbench.

Use without Blockbench

Paste this single instruction into Codex desktop app, Cursor, or another browser-capable agent:

Fetch and follow https://ashfox.io/workbench/agent-manifest.json using a direct HTTP request such as curl.

That manifest is the complete operating guide. It tells the agent how to open ashfox, connect to the page, inspect the project, edit safely, and ask what you want to create.

The workbench keeps its editable project in the browser. Files leave it only when the agent delivers the validated target artifact.

Editable demo gallery

The current featured set is Blackfrost Dreadwing, Aether Spear · Crownshift Interceptor, Ironroot · Verdant Crown Tractor, and Moonveil · Aurora Sovereign Kirin. Visit the demo gallery to inspect each animation and open the exact editable source directly in the workbench.

Every gallery entry is a self-contained folder under examples/gallery/. It contains one demo.json manifest, its media, and the exact project.ashfox source. The website discovers these folders automatically, exposes name search and category filters, and opens the selected archive directly in the workbench.

Automated gallery checks keep manifest metrics synchronized with each editable archive, audit connected eye and forward-facing foot anatomy, preserve every animation clip, and validate the exported GLB.

Local data and source

• The Blockbench plugin and MCP endpoint run on your machine.
• The web workbench does not require an ashfox account or project upload.
• The browser and Blockbench workflows are built and tested independently.
• Source for the plugin, sidecar, engine, workbench, and site is in this repository under the MIT license.

Build from source

git clone https://github.com/sigee-min/ashfox.git
cd ashfox
npm install
npm run build

The Blockbench build produces dist/ashfox.js and dist/ashfox-sidecar.js. To build or test one surface:

npm run build:blockbench
npm run build:public
npm test

See CONTRIBUTING.md for repository conventions and docs/ for user guides.

License

MIT. See LICENSE.