How to Use and Privately Deploy MCP Server for Chart Visualization

Overview

The mcp-server-chart project ( https://github.com/antvis/mcp-server-chart ) implements a Model Context Protocol (MCP) server that generates visual charts using the AntV ecosystem. It supports more than 25 chart types, ranging from basic line, bar, and pie charts to advanced visualizations such as Sankey, violin, and word‑cloud charts. The server can be integrated with any MCP‑compatible client (e.g., Claude, VSCode, Cursor).

Quick Start – Stdio Mode

Configure a local Stdio‑mode MCP server by adding the following JSON to the client configuration (use an absolute path for npx if necessary):

{
  "mcpServers": {
    "mcp-server-chart": {
      "command": "npx",
      "args": ["-y", "@antv/mcp-server-chart"]
    }
  }
}

For Windows, replace the command with cmd.exe and adjust the arguments:

{
  "mcpServers": {
    "mcp-server-chart": {
      "command": "cmd.exe",
      "args": ["/c", "npx", "-y", "@antv/mcp-server-chart"]
    }
  }
}

After configuring the server address in a client (e.g., Cherry Stdio) and issuing a query, the server returns a chart image URL such as

https://mdn.alipayobjects.com/one_clip/afts/img/qvkKTooCA1wAAAAASTAAAAgAoEACAQFr/original

. This URL is publicly accessible, which may be unsuitable for enterprise data.

SSE Transport Mode

Install the server globally: npm install -g @antv/mcp-server-chart Start the server with the desired transport:

# SSE transport (default endpoint: /sse)
mcp-server-chart --transport sse
# Streamable transport with custom endpoint
mcp-server-chart --transport streamable

Access the server at http://localhost:1122/sse or http://localhost:1122/mcp. The environment variable VIS_REQUEST_SERVER can be set to point to a private chart‑rendering service, enabling private chart generation.

Source Code Structure

Chart definitions : one file per chart type declares input parameters and tool descriptions.

Service folder : creates and starts server objects for each MCP service; serves as reusable templates for custom MCP servers.

Utils/generate.ts : calls remote rendering APIs and provides getVisRequestServer to retrieve the rendering endpoint dynamically.

Entry point : parses command‑line arguments and launches the appropriate MCP server based on the selected mode.

Private Deployment with Docker

Two Docker‑Compose files are used.

Chart rendering service (image ghcr.io/luler/gpt_vis_ssr:latest) exposing port 13100 and storing generated images in /app/public/images.

version: '3.8'
services:
  gpt-vis-ssr:
    image: ghcr.io/luler/gpt_vis_ssr:latest
    ports:
      - "13100:3000"
    volumes:
      - ./images:/app/public/images
    environment:
      - NODE_ENV=production
      - TZ=Asia/Shanghai
    restart: unless-stopped

MCP server – after cloning the repository, replace its docker-compose.yml with the following content (replace <YOUR_DOMAIN_OR_IP> with the actual host):

name: mcp-server-chart
services:
  sse:
    restart: unless-stopped
    build:
      context: ./sse
      dockerfile: Dockerfile
    ports:
      - "1123:1123"
    environment:
      - VIS_REQUEST_SERVER=http://<YOUR_DOMAIN_OR_IP>:13100/render
    volumes:
      - ./sse:/mcp-server-chart/sse

Start both containers. The private MCP server becomes reachable at http://<YOUR_DOMAIN_OR_IP>:1123/sse. Configure this address in an MCP client to use the privately hosted chart service.

Relevant discussion on private deployment can be found in the project issue tracker: https://github.com/antvis/mcp-server-chart/issues/140. Prompt templates for chart generation are available at https://github.com/antvis/GPT-Vis/blob/main/prompts/chart-type.md.