aseprite-mcp
A TypeScript MCP server that exposes Aseprite to LLM clients as 76 tools for drawing, layers, frames, palettes, and more by generating Lua scripts and running Aseprite headlessly.
README
aseprite-mcp
A TypeScript Aseprite MCP (Model Context Protocol) server.
It exposes Aseprite to LLM clients like Claude as 76 tools (drawing, layers, frames/cels, tilemaps, palettes, animation tweens, transforms, spritesheet export, ...), driving Aseprite headlessly by generating Lua scripts.
How it works
Each tool turns a high-level intent into a Lua script, then runs it with Aseprite -b --script <tmp.lua> and returns the result. The Lua is wrapped in xpcall, so runtime errors come back with a full traceback.
Requirements
- Node.js ≥ 20 (developed on Node 24)
- Aseprite (CLI-capable; Steam or standalone install both work)
Install
From npm (recommended):
npm install -g @letsagents/aseprite-mcp
After install, the aseprite-mcp command is available (see Run and the MCP client config below).
From source (development):
git clone https://github.com/letsagents/aseprite-mcp.git
cd aseprite-mcp
npm install
npm run build # tsc -> dist/
Locating the Aseprite executable
The Aseprite binary is resolved in this priority order:
- CLI flag
--aseprite-path <PATH> - Environment variable
ASEPRITE_PATH PATHsearch (aseprite/aseprite.exe)- Common Windows install locations (Program Files, Steam,
%LOCALAPPDATA%\Programs\Aseprite) - Error if none of the above is found
If your Aseprite is in a non-standard location (e.g. a Steam library on another drive), set it explicitly:
# Replace with the path to your Aseprite executable
export ASEPRITE_PATH="/path/to/your/Aseprite.exe"
Output paths
This server does not configure an output directory. Every filename argument must be an absolute path, provided by the caller (you, or the LLM after asking you). If the LLM doesn't know where to save, it asks you for the output location before invoking the tool. There is no --output-dir flag, no ASEPRITE_OUTPUT_DIR env var, and no default output directory.
Run
# After a global npm install:
aseprite-mcp
# From source:
npm start # node dist/index.js
# Or during development:
npm run dev # tsx src/index.ts (no build needed)
Connect your MCP client
Standard stdio server, supporting Claude Code / Cursor / Cline / Codex, etc. For Claude Code, .mcp.json:
{
"mcpServers": {
"aseprite": {
"type": "stdio",
// Use the bin name after a global npm install; from source use "node" + the absolute path to dist/index.js
"command": "aseprite-mcp",
"env": {
// Only needed when Aseprite isn't in a standard install location; otherwise the whole env block can be omitted
"ASEPRITE_PATH": "/path/to/your/Aseprite.exe"
}
}
}
}
Other clients (Cursor .cursor/mcp.json / Codex config.toml, etc.) use the same shape, just a different file location.
Scripts
| Script | What it does |
|---|---|
npm run build |
Compile src/ to dist/ (excludes tests) |
npm run dev |
Run directly via tsx (no build) |
npm start |
Run the compiled server |
npm test |
Run all unit tests (node:test, zero extra deps) |
npm run typecheck |
tsc --noEmit type check |
Tests
Tests use Node's built-in node:test runner — no test framework dependency. They are pure unit tests (assertions on generated Lua strings + handler logic via a mock runner); no test requires a real Aseprite install.
npm test
Project layout
src/
├── index.ts # entry: CLI args + Aseprite location + stdio transport
├── server.ts # McpServer: tools + 2 resources + 1 prompt
├── context.ts # Ctx { runner, verbose }
├── runner.ts # LuaRunner interface + RunOutcome
├── error.ts # AsepriteError + JSON-RPC error code mapping
├── aseprite/ # path resolution + subprocess spawn (xpcall wrapper)
├── lua/ # color parsing, Lua escaping, emit/* (per-domain Lua gen)
└── tools/ # per-domain tool registration (schema + handler)
Development guide
See AGENTS.md (cross-tool project instructions, shared by Codex / Cursor / Copilot / Gemini; architecture conventions + red lines + quick reference). The Claude Code entry is CLAUDE.md.
License
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。