Connect your AI to ycon.cc

ycon.cc speaks the Model Context Protocol. Point your AI assistant at one URL and it can search 300,000+ open-source icons and hand you ready-to-paste code for your framework.

https://ycon.cc/mcp

Streamable HTTP, JSON-RPC over POST. No authentication, no API key, no registration.

Add the server to your client

Pick your client and paste. Nothing else is needed.

Claude Code

claude mcp add --transport http ycon https://ycon.cc/mcp

Claude Desktop — claude_desktop_config.json

{
  "mcpServers": {
    "ycon": {
      "type": "http",
      "url": "https://ycon.cc/mcp"
    }
  }
}

Older Claude Desktop builds cannot speak HTTP directly. Use the mcp-remote bridge instead:

Claude Desktop — mcp-remote

{
  "mcpServers": {
    "ycon": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://ycon.cc/mcp"]
    }
  }
}

Cursor — .cursor/mcp.json

{
  "mcpServers": {
    "ycon": {
      "url": "https://ycon.cc/mcp"
    }
  }
}

VS Code — .vscode/mcp.json

{
  "servers": {
    "ycon": {
      "type": "http",
      "url": "https://ycon.cc/mcp"
    }
  }
}

curl

curl -sS -X POST https://ycon.cc/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

What your assistant can do

Seven tools. Your assistant picks between them on its own — you just describe the icon you want.

search_icons: Search icons by meaning and filter by style, set or licence. Returns identifiers, never markup, so it is cheap to call repeatedly.
get_icon_code: Install, import and usage snippets for one icon in any of the 13 supported frameworks, with its licence and attribution.
get_icon_svg: Raw SVG markup for one icon, optionally resized and recoloured.
preview_icons: Renders up to 12 candidates into a single image with a legend, so the model can see them before it chooses.
preview_icon: A larger rendering of one icon, for checking fine detail.
list_icon_sets: The icon sets available, with their prefixes, licences and icon counts.
list_icon_styles: The style values you can filter by — outline, solid, duotone and six more.

What a request looks like

Ask your assistant something like: Find me a shopping-cart icon for my React app, outline style, MIT licence. It will:

call search_icons with that query and those filters;
call preview_icons on the best candidates and actually look at them;
call get_icon_code for the winner and paste the React snippet into your project, licence noted.

Rate limits

The endpoint is public, so it is metered. The limits are set for an assistant working through a task, not for a crawler:

100 units of burst, refilled at one unit per second.
2,000 units per day. Most calls cost one unit; rendering previews costs more, because it starts a rasterizer.
Counted per IP address, and per /64 block for IPv6.

Every response carries X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset. A well-behaved client paces itself by reading them.

Over the limit you get a JSON-RPC error and a Retry-After header saying how many seconds to wait.

Licences

The icons are third-party open-source work and keep their own licences. Every answer carries the SPDX identifier and the attribution of the set the icon came from. Honouring them is up to you, not to us.

Browser-based clients are not supported

The endpoint sends no CORS headers and rejects cross-site requests, which is what makes it safe without an API key. Claude Code, Claude Desktop, Cursor and VS Code all connect from outside the browser, so this affects nobody using the configurations above.

What we record

We record anonymous usage statistics for each call — which tool was used, the search query, and the response time — to keep the service running. No API keys, no accounts, and IP addresses are anonymised.