MCP

Produce AntV charts in an assistant with generate_bar_chart

For desktop-assistant chart analysis with AntV after connecting via npx to @antv/mcp-server-chart; geographic maps currently cover China only.

Official source · MCP Server Chart

Install

npx -y @antv/mcp-server-chart

Who it is for

If you want AntV charts for data analysis inside a desktop assistant, MCP Server Chart fits. It is a TypeScript MCP service that produces charts through tools, and it can also be used in Dify.

It fits people who can already run npx. Official desktop-app configuration is the same set and names Claude, VSCode, Cline, Cherry Studio, and Cursor. Repository:

https://github.com/antvis/mcp-server-chart

It now supports 26+ chart types. If you have no npx and only want an offline local library, stop here — the main path is npx -y @antv/mcp-server-chart.

First try in 3 minutes

The official main path is desktop MCP configuration; do not install a global command first. Prerequisites: npx on this machine. Charts are generated in the local process; data enters that process.

Configuration is the client's mcpServers. Server name mcp-server-chart, command npx, arguments -y and @antv/mcp-server-chart:

npx -y @antv/mcp-server-chart

Low-risk first call: generate_bar_chart. Official notes say it compares numeric values across categories and fits side-by-side contrast. Pass only categories and numbers you already have; do not invent field names. This generates one chart; it does not change a remote database.

Observable success: a bar chart appears in the chat, or a chart you can open comes back. Official text does not write a return URL field. If the process does not start, first check that npx can pull @antv/mcp-server-chart.

Usage examples

Example 1: Category contrast. The tool is generate_bar_chart. Give it values for a few categories and compare them side by side. Required color or theme parameters are not written.

Example 2: See a trend under a continuous independent variable. The tool is generate_area_chart. Official text says it shows data trends under a continuous independent variable, so you can watch the overall direction. Likewise do not invent series names or axis fields.

Geographic chart tools use AMap from Amap and currently only support maps inside China. Do not add generate_* names that do not appear in the excerpt; the summary mentions maps, mind-map, and spreadsheet, but no parameter tables are given.

Limits

When HTTP transport is needed, official steps are first npm install -g @antv/mcp-server-chart, then pick a transport. The SSE default endpoint is /sse; Streamable can customize the endpoint. Access addresses on two separate lines:

http://localhost:1122/sse

http://localhost:1122/mcp

That is another path; do not mix it with npx stdio on the first try. Authentication is not written.

Geographic maps cover China only. Do not assume other countries or a world basemap. Unnamed clients are not verified. Data leaves your table and enters the local MCP process; this is not zero local read.