MCP

Render charts in chat with MCP Dashboards

For people who want interactive charts and KPIs inside MCP Apps-compatible clients, after installing only the official npm package mcp-dashboards and starting it with stdio or local HTTP.

Official source · MCP Dashboards

Install

npx -y mcp-dashboards --stdio

Who it is for

If you want interactive charts, dashboards, or KPI widgets in the chat itself instead of exporting data to another BI stack, MCP Dashboards fits. It offers 32 chart tools covering 45+ subtypes and renders in clients compatible with MCP Apps.

It fits people who can already run npx and whose client is on the official named set: Claude Desktop, Claude Web, VS Code (GitHub Copilot), Goose, Postman, MCPJam. ChatGPT is only written as rolling out; do not treat it as already usable. Repository:

https://github.com/KyuRish/mcp-dashboards

Install only the official npm package mcp-dashboards. Similarly named packages are not official; official text names mcp-dashboards-shared as having been malware. Package page:

https://www.npmjs.com/package/mcp-dashboards

First try in 3 minutes

The official main path is stdio. Prerequisites: Node/npx on this machine, and a client that can add an MCP server. Charts render in the chat; data enters the local process. Default HTTP mode binds to this machine only, but that is still not "no local read".

Install and configure in Claude's MCP entries. Run the official add command:

claude mcp add dashboard -- npx -y mcp-dashboards --stdio

The equivalent start argument is npx -y mcp-dashboards --stdio. Do not also install other similarly named packages.

Low-risk first call: after connecting, let the assistant use the built-in chart tools to draw a small set of numbers you already have as one chart or one KPI. The excerpt only writes "32 chart tools / 45+ subtypes"; it does not give tool names, fields, or sample data. So the first try only asks for an interactive chart in chat. Do not invent names like render_*, and do not invent a demo CSV URL.

Observable success: a chart, dashboard, or KPI widget appears in the client, and the process does not exit immediately. If you only see a text description, the current client may be outside the MCP Apps compatible set.

Usage examples

Example 1: Draw an interactive chart in chat. The capability is the 32 chart tools. Give the assistant categories or a time series you already have and let it pick a fitting subtype. Official text does not name the subtypes; success is only whether a chart appears.

Example 2: Fold the same numbers into a dashboard or KPI. The capability is dashboard and KPI widgets. This fits seeing several key metrics side by side instead of a single chart. There is likewise no layout parameter table; what you can confirm is only that it renders inside the chat.

If the client uses HTTP instead of stdio, the official command is npx mcp-dashboards, and the service appears at:

http://127.0.0.1:3001/mcp

By default it listens on this machine only. Do not expose it to the public internet — official text does not describe a reverse proxy or authentication.

Limits

The compatible set stops at the six clients above; ChatGPT is still rolling out. Do not assume an unnamed IDE can render MCP Apps.

The install surface is only the official package mcp-dashboards. Stop if you see mcp-dashboards-shared or another near name. stdio and local HTTP are the two process shapes that are written; no cloud hosted URL is written.

Official text does not give tool names, column constraints, themes, or an export interface. If something is missing, stop at "render inside the chat". Data stays in the local MCP process and the client session.