MCP Dashboards

Your AI can talk about data. Now it can show it.

The problem

We use AI for everything - analysis, reports, strategy. But when it comes to actually seeing the story in your data, you're stuck copying numbers into a spreadsheet and building charts yourself. The conversation has the insight. The visualization is somewhere else.

The solution

MCP Dashboards renders interactive charts, dashboards, and KPI widgets directly inside your AI conversation. 32 chart tools covering 45+ chart subtypes (bar has stacked/drilldown, hero has 11 variants, flowchart has 6 node kinds + 5 edge styles, etc.), 21 themes, 4 visual discovery catalogs, live polling, PNG/PPT/A4 export - all from a single MCP server. No browser tabs, no copy-paste, no context switching.

Quick Start

Claude Desktop

Add to your claude_desktop_config.json:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "dashboard": {
      "command": "npx",
      "args": ["-y", "mcp-dashboards", "--stdio"]
    }
  }
}

Claude Code / VS Code

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

Remote (Streamable HTTP)

npx mcp-dashboards
# Server starts on http://127.0.0.1:3001/mcp

Bound to localhost by default. See Configuration if you need to expose it on your network or allow browser access from a non-localhost origin.

Supported clients

Works in any MCP Apps-compatible client: Claude Desktop, Claude Web, VS Code (GitHub Copilot), Goose, Postman, MCPJam. ChatGPT support is rolling out.

Just ask

No API to learn. Describe what you want in plain English:

• "Compare Q1 vs Q2 revenue by region as a bar chart"
• "Show my portfolio allocation as a donut chart with the boardroom theme"
• "Build a dashboard with monthly sales KPIs and a trend line"
• "Visualize website traffic by country on a world map"
• "Track Bitcoin price live, updating every 30 seconds"
• "Show the conversion funnel from signup to purchase"
• "Rank the team by performance using a radar chart"
• "Diagram my multi-agent pipeline as a flowchart and let me step through it"

The AI picks the right tool, formats your data, and renders the chart inline. Click any data point to ask follow-up questions.

Don't know where to start? Ask "show me the catalog" - opens a master dashboard with one live visual tile per customization dimension (a real bar chart for charts, a pie for themes, a progress ring for hero variants, a neon-glowing card for effects). Click any tile, hit Ask, and you'll drill into that sub-catalog.

Interactive charts, not images

Every chart is interactive HTML rendered directly in your conversation:

• Explore in-chat - hover tooltips, click-to-select (feeds back to the AI), drill-down with breadcrumbs, scroll-zoom up to 12x on maps and heatmaps
• Editable flowcharts - drag nodes, multi-select, rename via right-click, minimap + node list to navigate, step-through playback of the flow, then sync your layout back to the AI to save it
• Live polling - real-time charts that auto-update from any API on a timer
• Export anywhere - PPT (16:9 slides), A4 (paginated with smart page breaks), PNG, CSV

Discovery / Catalogs - visual entry points for browsing what's available:

Typography is still configurable on every chart via typography=<name> (8 options listed under Themes) - it just doesn't have its own catalog because all 8 options were visually identical without per-card font loading, which wasn't worth the build for one preview tool.

Themes

21 built-in themes. Pass theme to any tool.

Mix-and-match with palette, typography (8 options: system, mono, professional, editorial, bold, techno, cyberpunk, luxury), and effects (5 presets: none, subtle, shimmer, neon, energetic).

Live Polling

Real-time charts that auto-update from any API. The live chart polls data via poll_http, which supports two modes:

Secure presets (authenticated APIs)

Configure presets via env vars. Credentials stay server-side and never appear in the conversation.

{
  "mcpServers": {
    "dashboard": {
      "command": "npx",
      "args": ["-y", "mcp-dashboards", "--stdio"],
      "env": {
        "POLL_PRESET_T212_CASH_URL": "https://live.trading212.com/api/v0/equity/account/cash",
        "POLL_PRESET_T212_CASH_HEADERS": "{\"Authorization\": \"Bearer YOUR_API_KEY\"}"
      }
    }
  }
}

Then ask: "Monitor my portfolio total and P/L live" - the AI uses render_live_chart with pollArgs: { preset: "t212_cash" }.

Naming: POLL_PRESET_<NAME>_URL and POLL_PRESET_<NAME>_HEADERS (JSON object).

Public URLs (no auth needed)

For public APIs, use the URL directly:

"Show me Bitcoin price updating every 30 seconds" - uses pollArgs: { url: "https://api.coingecko.com/api/v3/simple/price?ids=bitcoin&vs_currencies=usd" }.

How It Works

Built on MCP Apps. You ask the AI to visualize data, it calls the right tool, and the chart renders inline in your conversation. Self-contained - zero CDN, zero external requests.

MCP Apps supported (inline rendering): Claude Desktop, VS Code Insiders + MCP Apps extension, Goose, Postman.

No MCP Apps support? No problem. Every chart response also includes a clickable link to the same interactive chart in your browser:

• http://localhost:PORT/chart/{id} - the clickable link in chat. Backed by a tiny same-machine HTTP server bound to 127.0.0.1, lazy-started on the first preview request.
• file:///.../chart-{id}.html - the persistent shareable artifact. Same HTML bytes, written to disk so you can email, archive, or open it after the server has shut down.

Why both: Claude Code (VS Code) strips file://, vscode://, and command:// URL schemes from chat markdown, so only http:// produces a working clickable link. The file:// path is provided for sharing and offline viewing. Tool annotations (readOnlyHint, idempotentHint, openWorldHint) help clients reason about tool behavior.

Configuration

All optional. Defaults are safe - set these only if you need to override.

What's on disk

Chart HTML files are written to <system temp>/mcp-dashboards/ and auto-deleted after 7 days. The chart's built-in download button (PNG / PPT / A4) is the recommended way to save anything permanently.

Ask your AI to "list chart files" or "delete chart files older than 1 day" anytime - invokes list_chart_files / delete_chart_files.

Requirements: Node.js 18+.

git clone https://github.com/KyuRish/mcp-dashboards.git
cd mcp-dashboards
npm install
npm run build
npm run serve   # http://localhost:3001
npm run dev     # watch mode

Support

If MCP Dashboards is useful to you:

Built by Rishabh Dogra - connect on LinkedIn.

Privacy

All processing happens locally. No data is collected, transmitted, or stored externally.

External calls only happen when you explicitly ask - render_from_url and poll_http need a URL or preset you provide. Both refuse to fetch private, loopback, or link-local addresses (your 192.168.* network, AWS metadata at 169.254.169.254, etc.) so a prompt-injected AI can't quietly reach internal services. They also throttle per hostname so a runaway loop won't get your IP banned from a real API.

Credentials in env var presets never leave your machine. The browser preview server and the optional HTTP transport both bind to 127.0.0.1 by default and are not reachable from other devices. Chart HTML files live in your system temp folder and auto-delete after 7 days. See Configuration to change any of this.

License

FSL-1.1-MIT - Free to use for any purpose except building a competing commercial product. Each version converts to MIT two years after release. For commercial licensing, contact [email protected].

Third-party software

The bundled UI artifact (dist/mcp-app.html) embeds open-source libraries from the Chart.js, html2canvas-pro, Luxon, TopoJSON, and Dagre projects, distributed under MIT or ISC licenses. Full attribution and license texts are reproduced in THIRD_PARTY_LICENSES.md.