An MCP (Model Context Protocol) server that gives any AI assistant access to a collection of self-contained, browser-executable chart types. Each chart plugin is a single JSON file containing a renderScript — a JS function that loads its own CDN dependencies and renders into any DOM element.

The server runs on the Cloudflare Workers free plan (always-on, no cold-start sleep). After deploying your own (see Deploy your own), your endpoint is:

https://llm-chart-mcp.honganh32.workers.dev/mcp

Uses the current Streamable HTTP transport at /mcp. (The retired Railway deployment used the deprecated SSE transport at /sse.)

Deploy your own: See Deploy your own below.

Edit %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (Mac):

{
  "mcpServers": {
    "chart-plugins": {
      "type": "streamable-http",
      "url": "https://llm-chart-mcp.honganh32.workers.dev/mcp"
    }
  }
}

Restart Claude Desktop. The tools appear automatically.

claude mcp add chart-plugins --transport http https://llm-chart-mcp.honganh32.workers.dev/mcp

Add the Streamable HTTP URL https://llm-chart-mcp.honganh32.workers.dev/mcp in your client's MCP server settings.

pip install fastmcp
python mcp_server/server.py
# → http://localhost:8000/sse

• "I have sales data for 10 products over 4 quarters — suggest a chart and render it."
• "Show me how network connections between IP addresses evolved over 3 days."
• "Give me a gauge showing CPU utilization at 73%."

The AI will call suggest_chart, pick the best type, retrieve the renderScript via get_chart, and generate the code to render it in your app.

Each renderScript is a self-contained function(wrap, spec) string. Call it like this:

const plugin = await mcpClient.callTool("get_chart", { chart_type: "bar" });
const renderFn = new Function("return (" + plugin.renderScript + ")")();
renderFn(document.getElementById("my-chart"), {
  type: "bar",
  title: "Sales by Region",
  labels: ["North", "South", "East", "West"],
  datasets: [{ label: "Q1", data: [120, 85, 200, 60] }]
});

The function loads its own CDN scripts on first call and renders immediately. Multiple calls reuse already-loaded scripts.

A TypeScript port of the server lives in cloudflare_worker/. It runs on the Workers free plan with no credit card and no cold-start sleep.

cd cloudflare_worker
npm install
npx wrangler login
npm run deploy

Your endpoint is the printed URL + /mcp. Full guide: cloudflare_worker/README.md.

1. Create a Web Service pointing to this repo.
2. Build command: pip install -r mcp_server/requirements.txt
3. Start command: python mcp_server/server.py

Note: Render's free web services sleep after 15 minutes of inactivity and cold-start on the next request.

Want to add a chart type? See CONTRIBUTING.md.