Skip to main content
Glama

Stonecut Blockbench MCP

Website: https://stonecut.studio/resources | npm: https://www.npmjs.com/package/stonecut-blockbench-mcp | YouTube: https://www.youtube.com/@StonecutStudios (videos coming soon)

Stonecut Blockbench MCP is an MCP server and Blockbench plugin that lets AI agents like Claude Code build Minecraft models in the Blockbench desktop app. It makes GeckoLib mobs with bones, textures and animations, and Java item and block models with display transforms, while you watch every change and the agent checks its own work through screenshots. Free and open source under the MIT license.

It has two parts:

  • Stonecut Bridge, a Blockbench plugin (one readable file, stonecut_bridge.js).

  • stonecut-blockbench-mcp, an MCP server that your AI client starts.

The problem it solves

An AI agent cannot use Blockbench on its own, and writing model files by hand goes wrong in ways it cannot see: pivots in the wrong place, overlapping UVs, an item held at a strange angle. This tool gives the agent real Blockbench actions (add bones, add cubes, paint, animate, set display transforms, export) and shows it the result as images, so it can look, fix and try again like a person would. You keep control: it only touches the folders you allow, every edit can be undone in Blockbench, and a settings file decides what it may do.

Related MCP server: Blockbench MCP

Setup

  1. Copy the prompt in docs/setup-with-ai.md and paste it into your AI agent (Claude Code, or another agent that supports MCP servers). It checks that you have Node.js and Blockbench, shows its plan, and asks before it installs anything.

  2. When it tells you to, load the plugin in Blockbench: File > Plugins, click "Load Plugin from URL" and paste https://github.com/hitzerozero/stonecut-blockbench-mcp/releases/download/v1.0.0/stonecut_bridge.js, or download that file and use "Load Plugin from File" (see "Checking the plugin file" below). Click Allow on each permission question. You see "Stonecut Bridge listening on 127.0.0.1:38790".

  3. Ask it to build something, for example "Make a small cube creature with an idle animation in Blockbench and show me screenshots." A new tab opens in Blockbench and screenshots appear in the chat.

To set up without an agent, see "Manual setup" below. If something does not work, run npx -y stonecut-blockbench-mcp@1.0.0 doctor (see Troubleshooting).

Blockbench's official plugin list does not accept plugins that let an AI drive Blockbench, so Stonecut Bridge is shared from this GitHub repository instead of File > Plugins > Available.

Example requests

Copy one into Claude Code and change the details. Expect a few rounds of screenshots and notes.

  • Make a small slime mob with a 64 x 64 texture, an idle bounce and a hop animation. Show me screenshots from the front, side and iso.

  • Make a boss golem about three players tall, held to the quality bar in the Blockbench skill: layered stone plates, a glowing core, 128 x 128 texture, and idle, walk, two attacks, hurt and death animations. Show me the block-out before you paint.

  • Make a Java item model of a curved sword with a 32 x 32 texture. Set display transforms so it looks right in first person, third person and the inventory, and show me each of those.

  • Retexture the model that is open in Blockbench as a frozen version: ice blue tones, frost on the top faces, darker cracks. Keep the UV layout.

  • Add a heavy overhead slam attack to the open mob: a slow wind-up, a fast hit, a short shake, then recovery. Show me frames at the wind-up, the hit and the recovery.

  • Open output:wolf.geo.json with its animation file and tell me what bones and animations it has.

  • Export the open mob as geo, animation and texture into assets:geo, assets:animations and assets:textures/entity.

  • Make a crown as a Java item model for the head slot and tune the head display transform.

What it does

  • Opens new GeckoLib or Java item projects, or existing .geo.json and Java model .json files.

  • Adds bones with pivots and rotations, and cubes with box UV or per-face UV.

  • Creates and paints the texture: fill rectangles, set pixels, draw small pictures from a character grid, or import a PNG.

  • Adds animations and keyframes (rotation, position, scale).

  • Renders screenshots from named views or any angle, optionally mid-animation, and returns them to the agent as images.

  • Reads and sets Java display transforms (first person, third person, gui, ground and so on) and renders each display slot.

  • Exports .geo.json, .animation.json, Java model .json and texture .png into the folders you allow.

How it works

AI client (Claude Code) <--stdio--> stonecut-blockbench-mcp <--HTTP on 127.0.0.1--> Stonecut Bridge in Blockbench
  1. Your AI client starts the MCP server and calls its tools.

  2. The MCP server checks every call against a strict schema and against your settings file. Anything wrong is refused before it reaches Blockbench.

  3. It first makes the plugin prove that it holds the secret token (an answer to a random challenge), so another program on the port never sees the token. Then it sends the call with the token.

  4. The plugin listens on 127.0.0.1 only. It checks the token, refuses requests from web pages, checks the call again, keeps every path inside your output and assets folders, applies your settings, and asks you in Blockbench before replacing a file or closing a tab.

  5. The plugin makes the change in Blockbench as an undoable edit (Ctrl+Z works) and replies.

  6. The MCP server checks the shape of every reply before passing it to the agent. Screenshots come back as images.

Requests run one at a time. The plugin makes no outgoing connections.

Tools

Tool

What it does

Read-only or changes things

ping

Checks that Blockbench is open, the bridge is listening and the token is right.

read-only

status

GeckoLib plugin present, allowed folders, settings, open project, every open tab.

read-only

new_project

Opens a new tab with an empty GeckoLib entity model.

changes, adds only

open_model

Opens a .geo.json with its animation and texture, or switches to it if already open.

changes, adds only

close_tab

Closes duplicate tabs that have no edits.

changes, can replace

add_bones

Adds bones with pivots and rotations, parents first. All or nothing.

changes, adds only

add_cubes

Adds cubes with box UV offsets or per-face UVs. All or nothing.

changes, adds only

texture_create

Creates or resizes the texture, optionally filled with a colour.

changes, can replace

texture_fill_rect

Fills a rectangle of the texture with one colour.

changes, can replace

texture_set_pixels

Sets single pixels.

changes, can replace

texture_draw

Draws a small picture from rows of characters and a palette.

changes, can replace

texture_import

Loads a PNG from the allowed folders as the texture.

changes, can replace

add_animation

Adds an animation (loop, once or hold).

changes, adds only

add_keyframes

Adds rotation, position and scale keyframes to bones.

changes, adds only

screenshot

Renders views of the model and returns the images; writes PNGs to the output folder.

changes, can replace

export

Writes .geo.json, .animation.json, Java model .json and texture .png.

changes, can replace

open_java_model

Opens a Java block or item model .json, or switches to it if already open.

changes, adds only

new_java_item

Opens a new tab with an empty Java item model, optionally with display transforms.

changes, adds only

display_get

Reads the display transforms of the open Java model.

read-only

display_set

Sets display transforms within Minecraft's limits.

changes, can replace

display_screenshot

Renders display slots as Blockbench's display editor shows them and returns the images.

changes, can replace

"Changes, adds only" tools add to the project and never remove anything. "Changes, can replace" tools can overwrite pixels, transforms or, when asked with "overwrite": true, an existing file. Screenshot tools write a new PNG on every call and replace a file only when given a path and "overwrite": true. Each tool also reports these as MCP tool annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint; the last is false for every tool).

Paths are output:<relative path>, assets:<relative path>, or an absolute path inside one of those folders, for example output:shots/front.png. Colours are "#rrggbb", "#rrggbbaa" or "clear".

Security settings

The settings file is .stonecut_blockbench_config.json in your home folder. You own it; the plugin and the MCP server only read it. Create it with the defaults:

npx -y stonecut-blockbench-mcp@1.0.0 --init-config

Show where it is and the settings in use:

npx -y stonecut-blockbench-mcp@1.0.0 --config

The defaults, also in config.example.json:

{
  "output_folder": "",
  "assets_folder": "",
  "allow_export": true,
  "max_texture_size": 4096,
  "max_file_mb": 16,
  "max_cubes": 4096,
  "confirm": "overwrite_and_close"
}

Setting

Default

What it does

output_folder

"", meaning the Blockbench setting, or else stonecut-blockbench in your home folder

Folder for screenshots and exports. Readable and writable. Path prefix output:. A full path, or a path starting with ~ for your home folder.

assets_folder

"", meaning the Blockbench setting, or else off

Optional second folder, for example a resource pack or a mod's assets/<modid>. Readable and writable. Path prefix assets:.

allow_export

true

false refuses the export tool. Screenshots still go to the output folder, because that is how the agent sees its work, but only to new files with automatic names: no chosen path, nothing replaced.

max_texture_size

4096

Largest texture width or height in pixels, from 16 to 4096. Applies to new projects, texture_create and every PNG that is read.

max_file_mb

16

Largest model or texture file, in MB, that may be read or written, from 1 to 256.

max_cubes

4096

Most cubes one project may have, from 1 to 65536. Applies to add_cubes and to opening a model.

confirm

"overwrite_and_close"

When Blockbench asks you first. "overwrite_and_close": before replacing an existing file or closing a tab. "every_change": also before every tool that changes the project or writes a file, except screenshots that only add new PNGs. "never": no questions. If nobody answers within two minutes, the answer is no.

These are the only readable and writable places: the output folder and, if set, the assets folder. A non-empty folder in the settings file wins over the folder settings in Blockbench.

Fixed, and not changeable:

  • The bridge listens on 127.0.0.1 only and cannot be reached from other machines. The settings file refuses any host, bind, address, listen or interface key.

  • Every request must carry the secret token from .stonecut_blockbench_token in your home folder (32 random bytes, new every time the bridge starts). Before the server sends it, the bridge must prove it holds the token with an HMAC answer to a random challenge.

  • Requests from web pages (with an Origin header) or for another host name are refused.

  • Only .json and .png files, resolved so .. cannot leave the folders. Symbolic links and junctions inside the folders are not followed. Existing files are kept unless a call says "overwrite": true.

  • Texture references inside a Java model must be resource locations such as minecraft:item/iron_sword. A model whose textures name a file path or a data URL is refused, and a texture that is not found as a PNG inside the allowed folders shows blank.

  • One call can only ask for so much work: coordinates within -4096..4096, rotations within -360..360 degrees, inflate within -16..16, 4096 keyframes, 262144 characters in texture_draw (and 16777216 painted pixels), and 4096 bones in an opened model.

  • The plugin and the server are pinned: Claude Code runs stonecut-blockbench-mcp@1.0.0, and the plugin URL points at the file attached to the v1.0.0 release, so neither changes unless you update it.

  • No tool can read or write the settings file or the token file, and no tool runs code.

  • A settings file that is not valid makes the plugin and the server refuse every request until it is fixed. --config shows what is wrong.

Who can use it: any program running as your user account can read the token file, so the bridge trusts local programs as much as your account does. To report a security problem, see SECURITY.md.

Limits

  • Needs the Blockbench desktop app, open, with the plugin loaded. The web version cannot run it.

  • Screenshots use the live preview, so keep the Blockbench window open, not minimised.

  • Without the GeckoLib Blockbench plugin, new mobs are plain Bedrock entity models.

  • Tools add and change; they do not delete bones or cubes. Undo in Blockbench (Ctrl+Z) or delete by hand.

  • Java models: parent models are not followed, and textures outside the allowed folders show blank.

  • Files are .json and .png only. Save the .bbmodel yourself in Blockbench if you want one.

  • Confirm questions wait for you in Blockbench. While one is open, other requests wait.

  • Requests run one at a time; a call waits up to 3 minutes for Blockbench by default.

  • Not affiliated with Blockbench, GeckoLib or Mojang.

Troubleshooting

Run the doctor. It only reads and changes nothing:

npx -y stonecut-blockbench-mcp@1.0.0 doctor

From a source checkout: node dist/index.js doctor. Each line says PASS or FAIL, and every FAIL line says how to fix it. It checks:

  • the Node.js version (20 or newer);

  • the settings file: valid, missing (defaults in use) or broken, and which values are in use;

  • the token file;

  • whether the bridge answers on 127.0.0.1;

  • whether the plugin version matches the server version;

  • whether the GeckoLib plugin is installed in Blockbench;

  • whether the output folder exists and is writable.

Common messages:

  • "Bridge not reachable": Blockbench is closed, the plugin is not loaded, the bridge is stopped, or the port differs between the plugin setting and STONECUT_BB_PORT.

  • "No token at ...": the plugin has not started yet, or STONECUT_BB_TOKEN_FILE points elsewhere.

  • "Settings file ... has a bad value": fix the file; --config shows what is wrong.

  • "The person did not allow this": you clicked Deny, or nobody answered the confirm question in Blockbench within two minutes. See confirm under Security settings.

Remove it completely

  1. Run npx -y stonecut-blockbench-mcp@1.0.0 uninstall. It lists what it will remove and asks y/n: the server in Claude Code, the skill, the token file and the security settings file.

  2. In Blockbench: File > Plugins > Installed > Stonecut Bridge > Uninstall.

Your output folder (stonecut-blockbench in your home folder, or the one you chose) and your assets folder are kept on purpose, because they hold your models. Delete them by hand if you want.

Manual install details

Manual setup

The same setup without an agent:

  1. Install Node.js 20 or newer from https://nodejs.org (the LTS version). Check: node --version in a terminal prints v20 or higher.

  2. Install Blockbench from https://www.blockbench.net and open it. You see the Blockbench start screen.

  3. For mobs, in Blockbench: File > Plugins > Available, install "GeckoLib Models & Animations". It now shows under Installed.

  4. In Blockbench: File > Plugins, click "Load Plugin from URL" and paste https://github.com/hitzerozero/stonecut-blockbench-mcp/releases/download/v1.0.0/stonecut_bridge.js. Or download that file and use "Load Plugin from File".

  5. Blockbench asks to allow network access and access to a few files. Click Allow each time. You see "Stonecut Bridge listening on 127.0.0.1:38790".

  6. Install Claude Code if you do not have it (https://claude.com/claude-code), then run npx -y stonecut-blockbench-mcp@1.0.0 install in a terminal. It lists what it will do (add the server to Claude Code, copy the skill), asks y/n, and after y prints "Done." --dry-run only shows the plan.

  7. Run npx -y stonecut-blockbench-mcp@1.0.0 doctor with Blockbench open. Every line should say PASS.

  8. Start Claude Code and ask it to build something. A new tab opens in Blockbench and screenshots appear in the chat.

Adding it to Claude Code by hand

install runs these for you. To do it yourself:

  1. Add the server. macOS and Linux: claude mcp add --scope user stonecut-blockbench -- npx -y stonecut-blockbench-mcp@1.0.0. Windows: claude mcp add --scope user stonecut-blockbench -- cmd /c npx -y stonecut-blockbench-mcp@1.0.0. Claude Code answers that it added the server.

  2. Copy the skill (see "The skill" below).

  3. Start Claude Code and type /mcp. It lists stonecut-blockbench as connected.

npx -y stonecut-blockbench-mcp@1.0.0 uninstall removes the server, the skill, the token file and the settings file again. Both commands show what they will do and ask y/n first; add --yes to skip the question or --dry-run to only see the plan. Running install twice gives the same result.

Checking the plugin file

Each release lists the SHA-256 of stonecut_bridge.js; the repository has it in plugin/stonecut_bridge.js.sha256. Before "Load Plugin from File", compare it with the file you downloaded:

  • Windows (PowerShell): Get-FileHash stonecut_bridge.js -Algorithm SHA256

  • macOS: shasum -a 256 stonecut_bridge.js

  • Linux: sha256sum stonecut_bridge.js

Where the plugin file is

From npm, npx -y stonecut-blockbench-mcp@1.0.0 --plugin-path prints where the package keeps stonecut_bridge.js. Blockbench loads a file plugin from the same path at every start, so copy it somewhere it will stay before using "Load Plugin from File":

  • Windows (PowerShell): Copy-Item (npx -y stonecut-blockbench-mcp@1.0.0 --plugin-path) $HOME

  • macOS and Linux: cp "$(npx -y stonecut-blockbench-mcp@1.0.0 --plugin-path)" ~/

From source

git clone https://github.com/hitzerozero/stonecut-blockbench-mcp.git
cd stonecut-blockbench-mcp
npm install
node dist/index.js install

npm install also builds dist/index.js and dist/stonecut_bridge.js. From a source checkout, install points Claude Code at node <full path to the folder>/dist/index.js instead of npx. node dist/index.js uninstall and node dist/index.js doctor work the same way.

Other MCP clients

Run npx -y stonecut-blockbench-mcp@1.0.0 (on Windows cmd /c npx -y stonecut-blockbench-mcp@1.0.0) as a stdio server.

Plugin settings in Blockbench

File > Preferences > Settings > General, all starting with "Stonecut Bridge":

Setting

Default

What it does

port

38790

Port on 127.0.0.1. Stop and start the bridge after changing it.

output folder

stonecut-blockbench in your home folder

Used when the settings file does not set output_folder.

assets folder

empty (off)

Used when the settings file does not set assets_folder.

start automatically

on

Listen as soon as Blockbench opens. Tools > Stonecut Bridge: start / stop toggles it.

If you change the port, give the server the same one: claude mcp add --scope user stonecut-blockbench -e STONECUT_BB_PORT=38800 -- npx -y stonecut-blockbench-mcp@1.0.0.

Environment variables for the server: STONECUT_BB_PORT (default 38790), STONECUT_BB_TOKEN_FILE (default .stonecut_blockbench_token in your home folder), STONECUT_BB_TIMEOUT_MS (default 180000). --help lists them.

Blockbench asks for these permissions, and nothing else: network access to listen on 127.0.0.1, the token file, the settings file, and on first use your output folder and your assets folder if you set one.

The skill

skills/blockbench/SKILL.md tells an agent how to build mobs and items well and which rules to keep. For Claude Code, copy the skills/blockbench folder to ~/.claude/skills/ (all projects) or to .claude/skills/ in one project.

Check that it works

With Blockbench open and the plugin loaded, from the source folder:

npm run selftest

It opens new tabs (your open work is left alone), builds a cube mob and a plain sword, takes screenshots, exports into a new folder under <output folder>/selftest, reopens the exports and checks that bad requests are refused.

Development

npm install        # also builds dist/ and plugin/stonecut_bridge.js
npm test           # unit tests, no Blockbench needed
npm run typecheck
npm run build
npm run scan       # fails on personal paths, emails or ports in tracked files

Names you never want in the repository (your username, private project names) go one per line in .scan-names.local (git ignores it) or in STONECUT_SCAN_NAMES; the scan then refuses them too. The list itself is never stored in the repository.

The plugin source is split into small parts in plugin/src/. The build joins them, unminified, into dist/stonecut_bridge.js and plugin/stonecut_bridge.js; a test fails if the second is out of date. docs/plan.md describes the bridge protocol.

License

MIT, copyright Stonecut Studio. See LICENSE. Blockbench and GeckoLib belong to their authors.

Available Tools

21 tools
add_animationAdd an animationB

Adds an animation, for example idle, walk or attack. Looping animations should end on the pose they start with.

Inputs: name (required), loop, length.

add_bonesAdd bonesA

Adds bones (groups). Only bones animate, so every GeckoLib cube goes in a bone. Start with one root bone, then body, head, legs... with pivots where the joints are. Names must be unique. Nothing is created if any entry is bad.

Inputs: bones (required).

add_cubesAdd cubesA

Adds cubes. Give "from"+"to" or "origin"+"size". GeckoLib models use box UV: give each cube its own uv_offset so UV islands do not overlap. Java models may use "faces" for per-face UVs and need no bone. Nothing is created if any entry is bad.

Inputs: cubes (required).

add_keyframesAdd keyframesB

Adds keyframe tracks to an animation: per bone and channel (rotation in degrees, position in units, scale as factors).

Inputs: tracks (required), animation (required), interpolation.

close_tabClose duplicate tabsA
DestructiveIdempotent

Closes extra copies of a tab, keeping the first. A copy is closed only when it matches the kept tab (name, format, element and texture count) and has no edits, so no work is lost.

Inputs: name, duplicates.

display_getRead display transformsB
Read-onlyIdempotent

The display transforms of the open Java model: every set slot, or one slot.

Inputs: slot.

display_screenshotScreenshot display slotsA
Destructive

Renders display slots as the Blockbench display editor shows them and returns the images (PNG files also go to the output folder). First person is uncropped by default so you see how much of the view the item takes; gui is the inventory square.

Inputs: crop, path, slot, slots, +3 more.

display_setSet display transformsA
DestructiveIdempotent

Sets display transforms (undoable). Give "display" for several slots, or "slot" with rotation/translation/scale, or a "preset" file. Fields you leave out keep their value unless replace is true.

Inputs: slot, scale, preset, display, +3 more.

exportExport filesA
DestructiveIdempotent

Writes the open project: geo (.geo.json, GeckoLib/Bedrock), animation (.animation.json with all animations), model (Java block/item .json with its display block, Java projects only) and texture (.png). Existing files are kept unless overwrite is true.

Inputs: geo, model, texture, animation, +1 more.

new_java_itemNew Java item modelA

Opens a new tab with an empty Java block/item model, for example a weapon. Cubes need no bones. Optional display transforms, given directly or copied from a preset file.

Inputs: name (required), preset, display, namespace, +2 more.

new_projectNew GeckoLib modelA

Opens a new tab with an empty GeckoLib animated entity model (a Bedrock entity model if the GeckoLib Blockbench plugin is missing; the reply then has a warning). 16 units = 1 block; the model stands on y=0 and faces north (-Z).

Inputs: name (required), modid, texture_width, texture_height.

open_java_modelOpen a Java block/item modelA

Opens a Java block/item model .json that has "elements" in a new tab. Its texture comes from "texture", a same-named PNG next to the model, or assets/textures/...; anything else shows blank. Parent models are not followed. If the file is already open it switches to that tab.

Inputs: model (required), name, reload, reopen, +1 more.

open_modelOpen a .geo.json modelA

Opens an existing Bedrock/GeckoLib .geo.json in a new tab (converted to GeckoLib format when that plugin is installed), optionally with its .animation.json and texture PNG. If the file is already open it switches to that tab instead.

Inputs: geo (required), reload, reopen, texture, +1 more.

pingPing the bridgeA
Read-onlyIdempotent

Checks that Blockbench is open, the Stonecut Bridge plugin is listening and the token is right. Returns the plugin version.

Inputs: no named parameters.

screenshotScreenshot the modelA
Destructive

Renders the 3D preview from named views (or yaw/pitch in degrees) and returns the images; PNG files are also written to the output folder. The camera frames the model and is put back afterwards. "right" is the model's right side when it faces north. Give animation + time to see a pose. Screenshot often and check every side.

Inputs: yaw, path, time, view, +7 more.

statusBridge and project statusA
Read-onlyIdempotent

Whether the GeckoLib plugin is installed, the allowed folders (output, assets), the open project (format, bones, cubes, textures, animations, and for Java models the display transforms) and every open tab with its edit count. Call this first.

Inputs: no named parameters.

texture_createCreate or resize the textureA
DestructiveIdempotent

Creates the project texture, or resizes it (the canvas grows or shrinks and old pixels stay at the top-left). Also sets the project UV size unless set_uv_size is false.

Inputs: fill, name, width, height, +2 more.

texture_drawDraw a character gridB
DestructiveIdempotent

Draws a small picture from rows of characters, for example rows ["..ab..", ".abba."] with palette {"a": "#ff0000", "b": "#000000"}. Characters not in the palette are skipped. Each character covers scale x scale pixels.

Inputs: rows (required), palette (required), x, y, +1 more.

texture_fill_rectFill a rectangleC
DestructiveIdempotent

Fills a rectangle of the texture with one colour (or clears it).

Inputs: h (required), w (required), x (required), y (required), +1 more.

texture_importImport a PNG textureC
DestructiveIdempotent

Loads a PNG from the output or assets folder as the project texture.

Inputs: path (required), set_uv_size.

texture_set_pixelsSet pixelsC
DestructiveIdempotent

Sets single pixels: a list of [x, y, colour].

Inputs: pixels (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.4/5.0

    Scored across 21 tools

    Disambiguation4/5

    Most tools target a distinct resource+action: bones, cubes, animations, keyframes, textures, display transforms, screenshots, export. The only mild overlaps are the four project-creation openers (new_project, new_java_item, open_model, open_java_model) and the texture tools (create/import/set_pixels/draw/fill_rect), but the descriptions make the boundaries clear. screenshot vs display_screenshot is also well separated by purpose.

    Naming Consistency4/5

    Consistent snake_case throughout, mostly verb_noun or resource_verb (add_bones, texture_fill_rect, display_set, open_java_model). Minor deviations are the single-word verbs ping, status, screenshot and export, which are still unambiguous. The texture_* and display_* prefixes group related tools predictably.

    Tool Count4/5

    At 21 tools this is at the upper end, but the domain genuinely spans two model formats, texture painting, animation, display transforms and export, so nearly every tool earns its place. There is little redundancy; the count is heavy but not bloated.

    Completeness3/5

    The surface covers creation, opening, textures, animation, screenshots and export well, but it is add-only for geometry and animation: there is no way to move/resize/delete a cube or bone, remove a keyframe or animation, or read back texture pixels. An agent that makes a modeling mistake must restart the tab rather than fix it in place, which is a notable gap.

    Maintenance

    ActivityNo data
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers