SkillAgentSearch skills...

mascotforge

Give your AI agent an animated face. Describe a mascot, get a web component with idle, thinking, talking and error states. Blob and 3D styles are free; video uses your own fal key. CLI and MCP server for agents.

Install / Use

claude mcp add Arivex-ca -- npx -y github:Arivex-ca/mascotforge

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
🔌

MCP Server

Model Context Protocol server

Quality Score

83/100

Supported Platforms

Claude Code
Claude Desktop

Tags

Our assessment of mascotforge

mascotforge scores 83/100 on our quality scale, 2991st of 4,705 Development & Engineering skills we index.

Its MCP Server is 24 KB long, well organised into 23 sections with 13 code examples: a thorough specification that gives an agent plenty to work with.

It has 3 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
30/30
Structure
20/20
Description
15/15
Adoption
3/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated today, so mascotforge is actively maintained.
  • It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 92/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

Safety scan

Review

Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review judged it risky: The README instructs users to disable MCP tool approval for Codex exec by setting `default_tools_approval_mode = "approve"` and running `codex exec -c 'approval_policy="never"'`, which weakens agent safeguards.

AI review: risky

  • The README instructs users to disable MCP tool approval for Codex exec by setting `default_tools_approval_mode = "approve"` and running `codex exec -c 'approval_policy="never"'`, which weakens agent safeguards.
  • It is otherwise a standard project README documenting installation, API, and usage with no malicious commands or data exfiltration.
  • The install/run commands are user-facing documentation, not autonomous agent instructions.

AI review by kimi-k2.7-code on 2026-11-16. Automated pattern scan on 2026-11-16. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.

mascotforge compared with similar skills

All 4 of these similar skills score higher than mascotforge; compare them before choosing.

SkillScoreStarsUpdatedFormat
mascotforge (this skill)by Arivex-ca833todayMCP Server
taste-skillby Leonxlnx100117.0k8d agoCLAUDE.md
Agent-Reachby Panniantong100112.4k9d agoCLAUDE.md
rufloby ruvnet10081.9ktodayCLAUDE.md
headroomby headroomlabs-ai10079.9ktodayCLAUDE.md

Frequently asked questions

How do I install mascotforge?
Run claude mcp add Arivex-ca -- npx -y github:Arivex-ca/mascotforge. The install tabs above show the steps for each supported agent.
Which AI agents does mascotforge work with?
It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
Is mascotforge safe to use?
Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review judged it risky: The README instructs users to disable MCP tool approval for Codex exec by setting default_tools_approval_mode = "approve" and running codex exec -c 'approval_policy="never"', which weakens agent safeguards. It is MIT-licensed and scores 92/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is mascotforge still maintained?
The repository was last updated today, so mascotforge is actively maintained.
<p align="center"> <a href="https://arivex-ca.github.io/mascotforge/"><img src="docs/media/og.png" alt="MascotForge playground with a large animated agent mascot and live state controls" width="100%"></a> </p> <h1 align="center">MascotForge</h1> <p align="center"><b>Your users can't tell if your agent is thinking or stuck. Give it a face.</b><br> Design an animated status mascot, watch it follow your agent, and copy the web component. Blob and 3D styles run free in the browser.</p> <p align="center"> <a href="https://arivex-ca.github.io/mascotforge/"><b>Try it →</b></a> &nbsp;·&nbsp; <a href="https://arivex-ca.github.io/mascotforge/media/demo-video.mp4">Watch the video</a> &nbsp;·&nbsp; <a href="#for-agents">Docs</a> &nbsp;·&nbsp; <a href="README.fr.md">Français</a> </p> <p align="center"> <a href="https://arivex-ca.github.io/mascotforge/"><img src="docs/media/demo.gif" alt="Orbi follows a coding agent in a chat, then the same widget in five styles" width="800"></a> </p>

Use it for

  • A face for your chat assistant, with listening, thinking, and talking states.
  • A status indicator for a coding agent, from active work to an error or completed task.
  • Visible progress for background tasks, so users know the process is alive.
  • A responsive face for a voice agent, with state changes that need no mouth animation.

The first 10 mascots are on me. Tell me what your agent does and I'll make yours by hand, free. In return, three questions: would you install it, which states are missing, and what does your agent run on?

The studio, CLI help, and API errors are in French for now. Issues and pull requests in English are welcome: see CONTRIBUTING.md.

Gallery

MascotForge produced everything shown below.

AI video: a plush fox, built from one image that Claude Code generated, animated by Kling in four states: idle, thinking, talking, and error. The green background is removed and the loops are seamless. The preview places it on a cream background. Cost: $1.01 for four states.

The plush fox on video in idle, thinking, talking, and error states

Cursor tracking, shown here with the Arivex robot: an 8-second clip in which only the head looks in four directions. While idle, the widget displays the frame that looks toward the mouse.

The robot head looking left, right, up, and down

Code mascots: four Blob Atelier blobs and four Wobnimo characters, configured in the studio, each in a different state. They are free, animate in the browser, and follow the cursor.

Four blobs and four 3D characters in the MascotForge widget

Drawn by an agent: Codex designed this fox on its own over MCP, with no API key. After the first preview it decided the shape looked like a cat, then fixed it.

A fox blob designed by Codex over MCP, in eight states

Base image generated by the agent: no image to start from and no fal key? Claude Code generated these four plush characters itself, with the free Z-Image Turbo model through its Hugging Face connector, following the MascotForge skill. The chosen image then goes to the video, image-to-3D, or Wobnimo path.

Four plush mascots generated by Claude Code: fox, cloud, robot, and peach

3D character from an image: the source image on the left, followed by two Wobnimo characters written by a model from that image: Visor, then Glimmer with visual comparison. The result is an interpretation in Wobnimo's style, not a copy. It costs about $0.13 to $0.40 per character.

The robot image followed by two 3D interpretations

The same character in ten moods from four angles:

The Glimmer 3D character with angles, close-up, and moods

The studio

MascotForge studio home

Path selection in the studio

Final mascot integration

Key settings in the studio

For agents panel

MCP configuration for Codex CLI and the Codex app

Pipeline:

  1. Character: upload an image, preferably a cut-out PNG, or enter a text prompt. Flux generates the character on a solid green background. Uploaded images are placed on the same green.
  2. States on demand: start from a preset for AI agents, websites, or the full set, then add, remove, or freely describe up to 12 states. Each state is a 3-second clip that loops or plays once.
  3. Kling 3.0 image-to-video generates one clip per state. Each clip starts and ends on the source image through end_image_url. Loops stay seamless, and the widget switches states on the shared pose. Audio is disabled.
  4. ffmpeg keying removes the green. Each clip becomes a 480 px stacked-alpha video with color on top and the mask below.
  5. The <mascot-avatar> widget rebuilds transparency in WebGL across browsers, including Safari and iOS.
  6. Export produces an importable folder and a zip archive.

Requirements

  • Python 3.10 or newer
  • ffmpeg and ffprobe on the PATH, or MASCOTFORGE_FFMPEG. Generation still works without them, but mascots keep their green background. GET /api/health reports their availability.
  • For AI video only, a fal.ai key entered in the studio Settings. Blob and 3D mascots need no key. FAL_KEY remains a useful fallback for a local server. Without an available key, generation routes return 503 with a clear message.

Installation

git clone https://github.com/Arivex-ca/mascotforge.git
cd mascotforge
python -m venv .venv
.venv/bin/pip install -r requirements.txt   # Windows: .venv\Scripts\pip
cp .env.example .env                        # then set FAL_KEY

Run

.venv/bin/python -m uvicorn backend.main:app --port 8766 --env-file .env

--env-file loads FAL_KEY and local settings from .env. User-owned keys arrive through dedicated headers.

  • Studio: http://localhost:8766/
  • Widget and demo: http://localhost:8766/widget/

Key settings and security

The studio Settings button accepts a fal.ai key and an optional OpenRouter key. Without "Remember on this device," the studio uses sessionStorage. With the option enabled, it uses localStorage. All access to these stores falls back to memory if the browser blocks them.

The browser sends the required key in an X-Fal-Key or X-OpenRouter-Key header only when an operation needs it. The 3D image reader provider travels in X-Vision-Provider: fal|openrouter. Keys never enter a URL or query parameter. The server keeps a key in the request and task context. It does not write keys to jobs, data/, logs, or responses.

FAL_KEY is the local fallback when X-Fal-Key is missing. GET /api/health therefore continues to report only the server key. The studio displays the state of the browser key separately.

Do not set FAL_KEY on a server that other people can reach: every visitor without a key would generate videos on your fal account. The monthly cap (MASCOTFORGE_MONTHLY_CAP_USD) is also shared by everyone using the server.

The Test button starts no generation. OpenRouter checks the key with GET https://openrouter.ai/api/v1/key. fal.ai uses the fal-client storage token request without uploading a file or running inference.

For agents

The CLI and MCP server use the existing API. Configure the server address with MASCOTFORGE_URL, which defaults to http://localhost:8766. To provide keys without placing them in a URL or output, use MASCOTFORGE_FAL_KEY and MASCOTFORGE_OPENROUTER_KEY. The client sends them through the two dedicated headers.

CLI

python -m mascotforge styles
python -m mascotforge code --engine blob --set shape=cloud --out ./mon-blob
python -m mascotforge code --engine 3d --set character=volt --out ./mon-robot
python -m mascotforge blob guide --out ./guide-blob.md
python -m mascotforge blob check ./blob.json
python -m mascotforge blob preview ./blob.json --out ./planche-blob.png
python -m mascotforge blob export ./blob.json --out ./blob-sur-mesure
python -m mascotforge image3d ./robot.png --out ./robot-3d
python -m mascotforge video --image ./robot.png --usage agent --out ./robot-video
python -m mascotforge video --prompt "un petit renard orange" --usage site --out ./renard
python -m mascotforge video --image ./renard.png --states ./etats.json --out ./renard
python -m mascotforge status ID
python -m mascotforge export ID --out ./export
python -m mascotforge cost

Free loop: the agent draws the mascot itself

For a blob outside the eight built-in shapes, the agent describes a superellipse and its bumps in a JSON spec. blob check rejects intersections, outlines that are too thin, too small, or too large, and out-of-range values. blob preview displays all eight states before the standalone export. The guide includes a two-horned blob, a star, and a twisted bean.

A two-horned blob in eight states

A star blob in eight states

python -m mascotforge blob guide --out ./guide-blob.md
python -m mascotforge blob check ./blob.json
python -m mascotforge blob preview ./blob.json --out ./planche-blob.png
python -m mascotforge blob export ./blob.json --out ./blob-sur-mesure

An agent that can see an image can write the Wobnimo spec directly, without a fal.ai call or cost. The server retains the checker, renderer, Node runtime, and storage. The loop is visual:

  1. Inspect the reference image and read the guide.
  2. Write the JSON spec, then run the check.
  3. Generate the sheet and compare its front view in the top-left corner with the source image.
  4. Correct the spec, then repeat the check and preview.
  5. Save the spec or export it when the result is accurate.
python -m mascotforge spec guide --out ./guide-wobnimo.md
python -m mascotforge spec check ./personnage.json
python -m mascotforge spec preview ./personnage.json --out ./planche.png
python -m mascotforge spec save ./personnage.json
python -m mascotforge spec export ./personnage.json --out ./personnage-3d

These commands are free and do not require --yes. The image3d command is the paid option where a fal.ai model writes the spec.

Add --json to a command for agent-readable output. The image3d and video commands are paid. Without --yes, they display the estimated price and exit with a non-zero status before starting. Run the command again with --yes only after the user agrees.

MCP over stdio

python -m mascotforge mcp

Long-running creations use three calls: mascotforge_video_start or mascotforge_image3d_start, followed by mascotforge_status, then mascotforge_export. Paid start tools require user_confirmed=true. Their descriptions require the agent to state the price and obtain explicit consent.

The free mascotforge_spec_guide, mascotforge_spec_check, mascotforge_spec_preview, mascotforge_spec_save, and mascotforge_spec_export tools follow the same loop. The preview returns as an MCP image so the agent can inspect the sheet, compare it with the reference, and correct the spec before saving it.

The four custom Blob tools are mascotforge_blob_guide, mascotforge_blob_check, mascotforge_blob_preview, and mascotforge_blob_export. Their preview also returns as an MCP image.

Claude Desktop configuration in %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "mascotforge": {
      "command": "C:\\pa

Truncated for display — read the full file on GitHub.

Related Skills

View on GitHub
GitHub Stars3
CategoryDevelopment
Updated5h ago
Forks0

Languages

Python

Trust signals

92/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 low