MCP server

Connect Claude Code, Cursor, VS Code, or any MCP client to Vadivam so AI agents can search icons by tag or category and get SVG or framework code.

The Vadivam MCP server lets an AI agent find the right icon by name, synonym, or category, then get it as SVG or as an import for your framework. Agents stop guessing icon names that do not exist.

Use the hosted endpoint at https://vadivam.praveenjuge.com/mcp, or run the vadivam-mcp package locally. Both serve the same tools from the same icon data. The server is read-only and needs no account or API key.

Tools

Tool	Input	Returns
`search_icons`	`query`, optional `limit` (default 20, max 100), optional `category`	Ranked icons: exact name, then name prefix or word, then tag, then category
`suggest_icons`	`description`, optional `perConcept` (default 3, max 10)	Candidate icons for each thing a UI description mentions, plus the parts nothing matched
`get_icon`	`name`, optional `format`, `size`, `strokeWidth`, `color`, `absoluteStrokeWidth`	The icon as SVG, JSX, a framework snippet, data URI, base64, CSS mask, Tailwind class, CDN URL, sprite, font, Iconify id, or shadcn command
`list_categories`	none	Every category with its id, title, description, and icon count
`get_usage`	`framework`	Install command, import example, dynamic icon API, and docs link

format accepts svg, jsx, react, react-native, vue, svelte, solid, preact, astro, angular, data-uri, base64, css-mask, tailwind, cdn, sprite, font, iconify, and shadcn. suggest_icons splits a description such as “settings sidebar with search, profile, and logout” on commas and joins like “and” or “with”, searches each part, and ranks ties by Lucide’s popularity. An unknown icon name returns an error with the five closest real names. Tool results include text and structured content.

Connect to the hosted server

claude mcp add --transport http vadivam https://vadivam.praveenjuge.com/mcp

Add the server to .cursor/mcp.json in your project:

{
  "mcpServers": {
    "vadivam": {
      "url": "https://vadivam.praveenjuge.com/mcp"
    }
  }
}

Add the server to .vscode/mcp.json in your workspace:

{
  "servers": {
    "vadivam": {
      "type": "http",
      "url": "https://vadivam.praveenjuge.com/mcp"
    }
  }
}

Other clients connect to the same URL over Streamable HTTP.

Run it locally

The vadivam-mcp package runs the server over stdio. It bundles the icon data, so it works offline and matches the release you install. It needs Node.js 20 or newer.

claude mcp add vadivam -- npx -y vadivam-mcp

{
  "mcpServers": {
    "vadivam": {
      "command": "npx",
      "args": ["-y", "vadivam-mcp"]
    }
  }
}

{
  "servers": {
    "vadivam": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "vadivam-mcp"]
    }
  }
}

To host the server yourself, import the factory. createVadivamServer() returns a fresh McpServer, and createVadivamHttpHandler() wraps it in a stateless Streamable HTTP handler for Cloudflare Workers, Deno, or Bun:

import { createVadivamHttpHandler } from "vadivam-mcp";

export default createVadivamHttpHandler();

Agent skill

The agent skill teaches an agent how to install, import, size, label, and dynamically load Vadivam icons. It works with or without the MCP server. It also ships in the vadivam package at node_modules/vadivam/skills/vadivam-icons/SKILL.md, and is listed in https://vadivam.praveenjuge.com/.well-known/agent-skills/index.json. See AI agents.