fvi-mcp
MCP server for FreeVectorIcons design collections, enabling agents to search and fetch approved icons and assets instead of generating SVG from scratch.
README
fvi-mcp
MCP server for FreeVectorIcons design collections.
Connects Cursor, Claude Desktop, and other MCP clients to one FVI collection — icons, uploaded SVGs, strategy briefs, DESIGN.md, and related project files. Scoped to a single collection token.
npm: @freevectoricons/mcp
Why this exists
The usual agent workflow for icons is: prompt → inline SVG → paste into code. That breaks down quickly — inconsistent paths, missing viewBox, made-up attribution, different output on every retry.
FVI stores assets in a collection with stable ids, tags, and license fields. This server lets agents search and fetch what's already approved instead of generating new SVG from scratch.
It's a stdio MCP wrapper over the FVI HTTP API. Responses are metadata-first JSON; file bytes are fetched only when a tool asks for content or a download URL.
Setup
You need a FreeVectorIcons account and an MCP token for a design collection.
- Open a collection at freevectoricons.com
- Integrations → Setup MCP
- Create a token; copy the collection ID and secret
Add to .cursor/mcp.json (or your client's MCP config):
{
"mcpServers": {
"freevectoricons": {
"command": "npx",
"args": ["-y", "@freevectoricons/mcp"],
"env": {
"FVI_API_URL": "https://freevectoricons.com/api",
"FVI_COLLECTION_ID": "<collection-id>",
"FVI_TOKEN": "<collection-token>"
}
}
}
}
Use npx, not pnpm run. Package managers print lifecycle text to stdout, which breaks MCP stdio transport.
See mcp.json.example for a copy-paste template.
Environment variables
| Variable | Required | Default | Description |
|---|---|---|---|
FVI_TOKEN |
yes | — | Collection-scoped MCP token |
FVI_COLLECTION_ID |
yes | — | Collection id |
FVI_API_URL |
no | http://localhost:5001/api |
API base URL (include /api) |
For production: FVI_API_URL=https://freevectoricons.com/api
Tools
Read
| Tool | What it does |
|---|---|
get_design_collection_context |
Collection metadata, strategy brief, asset summary |
list_design_collection_assets |
All assets (metadata only) |
search_design_collection_assets |
Search by name, tag, category, or style |
get_design_asset |
One asset by id |
get_design_asset_content |
Inline UTF-8 for catalog SVGs, uploaded SVGs, Markdown |
get_design_asset_download_url |
Short-lived URL for PNG, PDF, and other binaries |
Write
| Tool | What it does |
|---|---|
create_design_asset_upload |
Create asset record + signed upload target |
park_design_asset |
Upload inline text or base64 into the collection |
create_design_asset_version |
Upload a new version of an existing asset |
Binary files are not embedded in MCP responses. Agents should use metadata + download URLs for images and PDFs.
Optional write metadata: drlProjectId, drlAssetType, documentType, tags.
Development
npm install
npm run build
node dist/mcp-server.mjs
To run from a local checkout in Cursor:
"args": ["-y", "file:/absolute/path/to/fvi-mcp"]
Build first (npm run build). The bundled entry is dist/mcp-server.mjs.
What this is not
- Not a global icon search API. One token, one collection.
- Not a guarantee of pixel-perfect icons. Catalog SVGs are AI-generated and refined on a schedule; check assets before production use.
- Not a replacement for your design review. It moves files and metadata; it doesn't judge brand fit.
Related
- freevectoricons.com/mcp — product docs
- ai-icon-generator — open pipeline that produces catalog icons
License
MIT — see 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。