icon-search-mcp
MCP server that finds open-source, commercially-usable icons and returns direct SVG download links, searching over 300,000 icons with license filtering.
README
icon-search-mcp
An MCP server that finds open-source, commercially-usable icons and returns direct SVG download links. Runs locally in Docker over stdio.
Icons come from the Iconify public API — 300,000+ icons across 231 open-source icon sets, each with machine-readable SPDX licence metadata.
Design rationale, recorded decisions, and known limitations live in docs/design.md.
How it works
Iconify's search index is keyword-based, not semantic — ?query=fast deployment returns
zero results. So the natural-language step lives in the calling model, not in this server:
User: "find me an icon that represents fast deployment"
│
▼ the host model translates the description into keywords
search_icons({ keywords: ["rocket", "deploy", "upload"] })
│
├─ one parallel Iconify query per keyword
├─ merge and dedupe, tracking which keywords each icon matched
├─ drop icons whose set fails the licence policy
└─ rank: most keywords matched first, then Iconify's relevance order
The tool description instructs the host model to do that translation and to retry with broader keywords when a search comes back empty.
Setup
docker build -t icon-search-mcp .
Then register it with your MCP client:
{
"mcpServers": {
"icon-search": {
"command": "docker",
"args": ["run", "--rm", "-i", "icon-search-mcp"]
}
}
}
For Claude Code: claude mcp add icon-search -- docker run --rm -i icon-search-mcp
Tool: search_icons
| Parameter | Type | Default | Description |
|---|---|---|---|
keywords |
string[] (1–5) |
required | Concrete English icon keywords, most relevant first |
limit |
number (1–64) |
12 |
Maximum icons to return |
license_policy |
enum | permissive |
See below |
Each result carries the icon id, set name, author, full licence details, and both a
download_url (serves Content-Disposition: attachment) and a preview_url.
{
"id": "material-symbols:rocket-launch",
"name": "rocket-launch",
"set": "Material Symbols",
"author": { "name": "Google", "url": "https://github.com/google/material-design-icons" },
"license": {
"spdx": "Apache-2.0",
"title": "Apache 2.0",
"url": "https://github.com/google/material-design-icons/blob/master/LICENSE",
"attribution_required": false,
"copyleft": false
},
"download_url": "https://api.iconify.design/material-symbols/rocket-launch.svg?download=1",
"preview_url": "https://api.iconify.design/material-symbols/rocket-launch.svg",
"matched_keywords": ["rocket", "launch"]
}
Licence policies
| Policy | Allows | Reach |
|---|---|---|
permissive (default) |
MIT, Apache-2.0, ISC, BSD-3-Clause, CC0-1.0, Unlicense, OFL-1.1, MPL-2.0 | ~224k icons |
permissive_plus_attribution |
above + CC-BY-4.0, CC-BY-3.0 — you must credit the author | ~291k icons |
all |
above + copyleft (GPL-*, CC-BY-SA-*) — share-alike terms apply to derivatives | ~315k icons |
NonCommercial sets (cbi, ps) are excluded under every policy, including all.
Licence data is reported as published by each icon set; verify it against the linked LICENSE file before shipping an icon in a product. This server surfaces licence terms, it does not give legal advice.
Development
npm install
npm test # builds, then runs offline unit tests against fixtures
npm start # run the server directly on stdio, without Docker
src/licenses.ts and src/search.ts are pure functions with no I/O; src/iconify.ts is
the only module that touches the network.
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。