figma-meta-mcp
Figma MCP server that lets agents execute arbitrary JavaScript inside a running Figma desktop app via a local development plugin, enabling real-time document inspection and manipulation while avoiding REST API rate limits.
README
figma-meta-mcp
Lightweight Figma MCP in the same mental model as
ae-meta-mcp (ae_exec) and
CocosMetaMCP (cocosmcp_exec): the Cursor agent runs JavaScript inside a
running Figma via figma_exec.
Cursor Agent ──stdio MCP──▶ Node (MCP + bridge :3851)
▲
│ WebSocket /plugin
Figma Development plugin (UI + main)
│
figma.* Plugin API
This path does not use the Figma REST API + Personal Access Token, so it avoids the Viewer-seat REST monthly rate limits. You need the Figma desktop app and the local Development plugin panel open.
Repo: shinjiyu/figma_mcp
Tools (MVP)
| Tool | Description |
|---|---|
figma_health |
Bridge up? Plugin WebSocket connected? |
figma_exec |
Run arbitrary Plugin-API JS in the main thread; return JSON |
figma_selection_info |
Summarize current page + selection bounds |
Requirements
- Windows / macOS / Linux
- Node.js >= 18
- Figma desktop app
- Cursor (or any MCP client)
Install
git clone https://github.com/shinjiyu/figma_mcp.git
cd figma_mcp
npm install
Wire Cursor
npm run setup:cursor
Paste the printed snippet into ~/.cursor/mcp.json, then toggle figma-meta-mcp
off/on under Cursor Settings → MCP.
Example:
{
"mcpServers": {
"figma-meta-mcp": {
"command": "node",
"args": ["D:/workspace/figma_mcp/mcp/index.mjs"],
"env": {
"FIGMA_MCP_PORT": "3851"
}
}
}
}
Import the Figma plugin
- Start Cursor so MCP (and the bridge on
127.0.0.1:3851) is running. - Open your design file in the Figma desktop app.
- Menu: Plugins → Development → Import plugin from manifest…
- Select
plugin/manifest.jsonfrom this repo. - Run Plugins → Development → figma-meta-mcp and keep the panel open.
The panel should show Connected · plugin channel ready.
Verify
figma_health→{ ok: true, pluginConnected: true }figma_selection_info→ current page + selectionfigma_exec:
return {
file: figma.root.name,
page: figma.currentPage.name,
count: figma.currentPage.children.length,
};
Writing figma_exec code
- Runs in the plugin main thread with global
figma. - Write an async body; use
returnfor the payload. - Return plain JSON (id / name / type / x / y / width / height). Live node proxies are auto-shrunk when possible, but prefer mapping yourself.
- Need document bytes for a node?
await figma.getNodeByIdAsync(id)(dynamic-page).
See skills/figma-plugin-api/SKILL.md.
Layout
mcp/ stdio MCP server (index, core, context)
bridge/ HTTP + WebSocket host used by the plugin UI
plugin/ Figma Development plugin (manifest, code.js, ui.html)
scripts/ setup-cursor.mjs
skills/ agent notes for Plugin API
examples/ cursor-mcp.json
Env
| Variable | Default | Meaning |
|---|---|---|
FIGMA_MCP_PORT |
3851 |
Bridge listen port |
FIGMA_MCP_HOST |
(unset) | Default: bind loopback only (127.0.0.1 + ::1). Set only to override. |
FIGMA_MCP_CORS_ORIGINS |
Figma + null + localhost |
Comma-separated Origin allowlist (not *) |
FIGMA_MCP_EXEC_TIMEOUT_MS |
30000 |
Per-job timeout |
Not in scope (yet)
- Official remote Figma MCP OAuth (
mcp.figma.com) — use that separately if you want cloud tools - REST PAT wrappers — intentionally avoided for quota reasons
- Recipe/promote layer (CocosMetaMCP-style) — add later if needed
License
MIT
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。