Illustrator AI MCP Server
An MCP server that lets AI assistants control Adobe Illustrator locally—running ExtendScript, capturing canvas screenshots, exporting artwork, and optionally remembering techniques—all without network activity.
README
illustrator-ai-mcp
Let Claude design in Adobe Illustrator — 100% local, private by default.
An MCP (Model Context Protocol) server that gives AI assistants hands and eyes inside Adobe Illustrator: it writes and runs ExtendScript, sees the canvas via window screenshots, reads document structure, exports artwork, and (optionally) remembers techniques that worked — all without a single byte leaving your machine from this server.
📖 New here? Start with the step-by-step USER_GUIDE.md — install, Claude Desktop/Code setup, permissions, and troubleshooting for both platforms.
| Platform | Status |
|---|---|
| 🍎 macOS | ✅ Fully supported & tested (AppleScript bridge) |
| 🪟 Windows | ⚠️ Experimental (COM bridge via pywin32 — testers welcome) |
Privacy
- No network code. This package contains zero HTTP clients, telemetry, or analytics. It talks only to Illustrator (via AppleScript/ExtendScript) and your local filesystem.
- History is opt-in and OFF by default. Nothing is logged until you (or the AI, with your permission) call
set_memory(True). - One-command wipe. Everything the server ever stores lives in
~/.illustrator-mcp. Delete that folder and it's gone. There is also aclear_historycapability in the memory layer. - Honest caveat: screenshots of your Illustrator window and script results are sent to the AI model as part of your conversation — that is inherent to how the AI sees the canvas and iterates on the design. If a document is confidential, don't ask an AI to look at it. The server itself, however, never transmits anything on its own.
Features vs. the original
Inspired by spencerhhubert/illustrator-mcp-server, rebuilt with the feedback loop an AI actually needs:
| Capability | spencerhhubert/illustrator-mcp-server | illustrator-ai-mcp |
|---|---|---|
| Run ExtendScript in Illustrator | Yes | Yes |
Script output capture (return values) |
No — fire and forget | Yes — scripts return data to the AI |
| Error messages with line numbers | No | Yes — AI can fix and retry |
| Window screenshots | Fixed screen region | Dynamic window detection, any display |
| Document state as JSON (artboards/layers/selection) | No | Yes |
| Export artwork (png/svg/pdf/jpg) | No | Yes |
| Undo | No | Yes |
| Local run history | No | Yes — opt-in, off by default |
| Reusable snippet library | No | Yes — searchable across sessions |
Requirements
- macOS (AppleScript bridge — fully supported) or Windows (COM bridge — experimental)
- Adobe Illustrator (any recent CC version), running
- Python 3.11+
- uv recommended (plain pip works too)
Install
With uv (recommended)
git clone https://github.com/Gokul2/illustrator-mcp-Gokul-Dhatkshina-Moorthy-.git illustrator-ai-mcp
cd illustrator-ai-mcp
uv sync
With plain pip
git clone https://github.com/Gokul2/illustrator-mcp-Gokul-Dhatkshina-Moorthy-.git illustrator-ai-mcp
cd illustrator-ai-mcp
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
Hook it up
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"illustrator": {
"command": "uv",
"args": [
"--directory",
"/absolute/path/to/illustrator-ai-mcp",
"run",
"illustrator-ai-mcp"
]
}
}
}
Restart Claude Desktop afterwards.
Claude Code
claude mcp add illustrator -- uv --directory /absolute/path/to/illustrator-ai-mcp run illustrator-ai-mcp
First run: macOS permissions
The first time the server touches Illustrator, macOS will ask for two permissions. Grant both to the app hosting the server (Claude Desktop, your terminal, etc.):
- Automation — lets the server send ExtendScript to Illustrator. System Settings → Privacy & Security → Automation → enable Adobe Illustrator under your client app.
- Screen Recording — lets
view_canvasscreenshot the Illustrator window. System Settings → Privacy & Security → Screen Recording → enable your client app, then restart it.
If a screenshot comes back black or the tool reports a permission error, Screen Recording is the usual culprit.
Tool reference
| Tool | What it does |
|---|---|
run_script(code) |
Execute ExtendScript on the current document. return <value> sends data back; errors include line numbers for fix-and-retry. |
view_canvas() |
JPEG screenshot of the Illustrator window — the AI's eyes. |
get_document_state() |
Structured JSON of artboards, layers, and selection — cheaper than a screenshot for layout logic. |
export_artwork(path, format) |
Export the document to png / svg / pdf / jpg at a path you choose. |
undo_last() |
Undo the most recent action (Cmd+Z). |
list_history(limit) |
Recent script runs (only when memory is enabled). |
get_run(run_id) |
Full code + screenshot path of one past run. |
save_snippet(name, description, code, tags) |
Save a technique that worked to the local library. |
search_snippets(query, limit) |
Search the library before writing complex scripts from scratch. |
set_memory(enabled) |
Opt in/out of local history logging (default: off). |
memory_stats() |
What's stored, where, and a privacy statement. |
Example prompts
- "Design a minimalist mountain logo — dark blue palette, geometric, on a square artboard."
- "Recreate this layout as 3 variations on separate artboards."
- "Draw a sunburst badge with 24 rays, then export it as an SVG to my Desktop."
- "Look at the canvas and tell me what's misaligned, then fix it."
- "That gradient trick worked great — save it as a snippet for next time."
Your data
Everything the server stores lives in one folder:
~/.illustrator-mcp/
That's the opt-in run history (SQLite), optional screenshots, and your snippet library. To wipe it:
- ask the AI to clear the history (uses the memory layer's
clear_history), or - nuke it yourself:
rm -rf ~/.illustrator-mcp
Nothing else is written anywhere, and nothing is ever transmitted by this server.
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。