animation-mcp
Enables AI coding assistants to create and refine Roblox combat animations through tool calls, supporting the full pipeline from motion composition to Roblox Studio import.
README
animation-mcp
Animation MCP lets an AI coding assistant create Roblox combat animations by calling tools instead of hand-editing bones, CFrames, or rig math.
If you are new to MCP: think of this project as a local animation workstation for your AI app. You connect it once in Claude Code, Codex, Cursor, or another MCP client, then ask the AI for moves like "make a fast anime 4-hit M1 combo." The AI calls this server's tools, previews and critiques the result, repairs any mechanical issues, and exports Roblox-ready files.
One-command install
On Windows, open PowerShell in this folder and run:
.\install.ps1
Or double-click/run:
install.cmd
That single command installs dependencies, refreshes project-local MCP configs,
registers Codex and Cursor user configs, tries Claude Code registration when the
claude CLI is available, and runs the setup doctor.
To preview what it would change:
.\install.ps1 -DryRun
irm remote command:
$env:ANIMMCP_REPO_ZIP="https://github.com/ThatGuyTHD/animation-mcp/archive/refs/heads/main.zip"; irm "https://raw.githubusercontent.com/ThatGuyTHD/animation-mcp/main/install.ps1" | iex
More installer options are in docs/INSTALLER.md.
Manual 5-minute setup
- Install Python 3.10+ and uv.
- Open a terminal in this folder.
- Install the Python environment:
uv sync
- Check that the project is healthy:
uv run python scripts/doctor.py
- Print copy/paste config for your AI app:
uv run python scripts/doctor.py --configs
That command prints ready-to-use snippets for Claude Code, Codex, Cursor, and generic MCP clients using the exact path to this checkout.
Connect your AI app
Claude Code
Run this from any terminal, replacing the path if needed:
claude mcp add animation-mcp --scope user -- uv --directory "C:\path\to\Animation MCP" run animation-mcp
claude mcp list
This repo also includes a project-scoped .mcp.json for Claude Code. If you open
Claude Code from this project folder, Claude can use that local config instead.
Codex
Add this to ~/.codex/config.toml, or keep the project-scoped
.codex/config.toml in this repo:
[mcp_servers.animation-mcp]
command = "uv"
args = ["--directory", "C:/path/to/Animation MCP", "run", "animation-mcp"]
startup_timeout_sec = 30
tool_timeout_sec = 300
[mcp_servers.animation-mcp.env]
ANIMMCP_ENGINE = "auto"
Cursor
Cursor can use the included .cursor/mcp.json when this folder is open. For a
global install, add this to your Cursor MCP config:
{
"mcpServers": {
"animation-mcp": {
"command": "uv",
"args": ["--directory", "C:/path/to/Animation MCP", "run", "animation-mcp"],
"env": {
"ANIMMCP_ENGINE": "auto"
}
}
}
}
After connecting, open your client's MCP tools view and confirm that
animation-mcp exposes tools such as create_rig_session, compose_move,
solve_animation, render_preview, critique_motion, and
export_for_roblox.
More complete setup paths live in docs/GETTING_STARTED.md and docs/INTEGRATIONS.md.
First prompt to try
Paste this into your AI app after the MCP is connected:
Use animation-mcp to create a fast anime 4-hit Roblox M1 combo for an R15 blocky
rig. Read the style bible and quality gates first. Solve it, render a contact
sheet, critique it, repair mechanical issues if needed, and export Roblox files.
The usual AI workflow is:
create_rig_session
read animmcp://docs/style-bible and animmcp://docs/quality-gates
search_motion_reference / list_pose_library
compose_move
refine with IK, root motion, hit pause, markers, and style modifiers
solve_animation
render_preview
critique_motion
repair_animation if needed
export_for_roblox
Copy/paste workflow prompts are in docs/AI_WORKFLOWS.md.
Requirements
- Python 3.10+ and
uvrecommended - Optional: Blender 4.x/5.x for rendered previews, MP4, FBX, and
.blendexports. Without Blender, the built-in pure-Python solver still creates animation data and stick-figure contact sheets/GIFs. - Optional: Roblox Studio plus the companion plugin in roblox/AnimMCPImporter.luau
Optional environment variables:
| Var | Meaning | Default |
|---|---|---|
ANIMMCP_ENGINE |
auto, blender, or fallback |
auto |
ANIMMCP_BLENDER_PATH |
Explicit blender.exe path |
auto-detected |
ANIMMCP_EXPORT_DIR |
Output directory | ./exports |
ANIMMCP_ASSETS |
Asset library root | ./assets |
Roblox Studio import
- Save roblox/AnimMCPImporter.luau as a local Studio plugin.
- Enable Allow HTTP Requests in Game Settings > Security.
- Ask your AI to run
export_for_roblox(serve=true). - In Studio, use Import Animation to pull
anim.jsonfrom the local server. - Use Register Temp Id for quick playtesting, and Build CombatConfig Module for hit windows, damage, knockback, cancel windows, SFX, VFX, and camera shake data.
Outputs land in exports/<move_name>/. When serving is enabled, the Studio
plugin reads from http://127.0.0.1:8973/latest/anim.json.
Verify the full pipeline
uv run python scripts/smoke_test.py
uv run python scripts/smoke_test.py --blender
The first command uses the pure-Python fallback engine. The second also checks Blender rendering and FBX export when Blender is installed.
What this MCP gives the AI
- Semantic animation tools: choreography, poses, fragments, IK targets, root motion, hit pause, markers, timing, and style
- Resources the AI can read: style bible, combat timing, marker vocabulary, export rules, quality gates, bad examples, rigs, poses, grammar, and references
- Quality loop: solve, render, critique, auto-repair mechanical issues, export
- Roblox outputs:
anim.json, marker sidecar, combat config JSON, Luau module, and optional FBX/Blend files through Blender
See DESIGN.md for the architecture.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。