What is the MCP UI Components server?
MCP UI Components is an open-source Express and React server that gives AI chat clients widget-based output: basic text, radio and checkbox groups, a line chart, an e-commerce product list, a QR code, and a data table. It is built on the Flowbite UI framework and the SkyBridge framework, and it exposes an MCP endpoint a client reaches through a tunnel.
MCP UI Components is an open-source library of UI widgets for MCP applications, built on the Flowbite UI framework together with the SkyBridge framework. Instead of a chat client showing a tool's raw JSON result, the server renders that result as an interactive widget.
It ships as an Express server with an MCP endpoint plus a React application for the widgets themselves, so the same project both serves the protocol and renders what a client displays.
What it changes in your workflow
- A tool result that would otherwise print as raw JSON instead renders as a line chart, a data table, or a selection prompt the person can interact with.
- Radio and checkbox group widgets turn a decision point in a conversation into a clickable prompt instead of a typed answer.
- The project supports several predefined themes, including default, minimal, enterprise, playful and mono, so the rendered widgets can match a brand's look.
- Because the server runs locally and is reached through a tunnel, a developer can iterate on a new widget and see it rendered in a real client quickly.
Setup at a glance
| MCP UI Components server | |
|---|---|
| Local start | npm install, then npm run dev on port 3000 |
| Remote reach | A tunnel such as ngrok, with /mcp appended to the forwarding address |
| Node version | Node.js 24 or newer |
| Clients named | ChatGPT, Claude Web, Claude Code, Gemini CLI, Cursor, VS Code, Mistral AI, Windsurf, Codex |
Who it is for
Best for
- Teams building an MCP app that wants its tool output to render as a widget instead of plain text or JSON.
- Prototyping a selection prompt, chart or table widget before building a custom one from scratch.
- Developers already using Flowbite and Tailwind who want matching widgets inside a chat client.
Not for
- If the deliverable is production React chart code for a real web app rather than a chat widget, the Data Visualization Mastery skill picks from over 50 chart types with real library code.
- When only visualization design guidance is needed rather than a rendered widget, Generative UI MCP loads chart and diagram design rules on demand.
- Go with a WordPress-page route when the task is editing a live Elementor page rather than a chat widget, MCP Abilities Elementor inspects and edits a page's own widgets instead.
- Choose a hosted design-system server when charts and components should come from Carbon rather than a chat widget set, Carbon MCP gives direct access to Carbon's own charts and components.
Use it with UiChemy
This server renders widgets inside a chat client, a different surface from a web page. UiChemy's Composer builds the WordPress page itself inside Elementor or Gutenberg, so the two cover separate parts of a product's interface.
Connect MCP UI Components Server by Flowbite to your AI client
git clone https://github.com/themesberg/mcp-ui-components.gitClone and install
Run
git clone https://github.com/themesberg/mcp-ui-components.git, thennpm installinside the project.Start the server
Run
npm run devto launch the Express server with its MCP endpoint and the React widget app, on port 3000 by default.Open a tunnel
Run
ngrok http 3000and take the forwarding address ngrok prints.Register the endpoint
Append /mcp to the forwarding address and register that as the server's URL in your chat client.
Try a widget
Ask the connected client to call a tool that returns a chart, table or selection prompt and confirm it renders as a widget, not raw text.
Client configuration
{
"mcpServers": {
"mcp-ui-components": {
"url": "https://<your-ngrok-subdomain>.ngrok.app/mcp"
}
}
}Endpoints and ports
npm run devLaunches the Express server and MCP endpoint on port 3000.ngrok http 3000Exposes the local server at a public forwarding address for a remote client.What you can do with it
- Widget output instead of JSONTool results render as an interactive widget in the client interface instead of a raw JSON block.
- Seven ready widgetsText, radio and checkbox groups, a line chart, a product list, a QR code and a data table cover common output shapes.
- Built on Flowbite and TailwindWidgets are styled with the Flowbite UI framework and Tailwind CSS rather than a one-off custom styling layer.
- Several predefined themesDefault, minimal, enterprise, playful and mono themes let the rendered widgets match a brand's look.
- Wide client coverageThe project names ChatGPT, Gemini CLI, Cursor, VS Code, Mistral AI, Windsurf, Codex and other MCP-compatible agent clients as compatible.
- Local-first developmentThe server runs locally by default, with a tunnel such as ngrok used only to expose it for a remote client to reach.
Tools it exposes
Basic Text
Renders a simple text widget instead of plain string output.
Radio Group
Renders a single-selection prompt the person clicks instead of typing an answer.
Checkbox Group
Renders a multi-selection prompt for picking more than one option.
Line Chart
Renders a data visualization widget for a series of values.
E-commerce Products
Renders a product listing widget with images and details.
QR Code
Renders a generated QR code as a widget instead of a text string.
Table Data
Renders structured rows and columns as an interactive table widget.
Frequently asked questions
What is the MCP UI Components server?
It is an open-source server by Flowbite that renders MCP tool output as interactive widgets, such as a chart, table or selection prompt, instead of a raw JSON block in a chat client.
How do I run the MCP UI Components server?
Clone the repository, run npm install, then npm run dev to start the Express server with its MCP endpoint on port 3000. Open a tunnel such as ngrok and append /mcp to the forwarding address to reach it from a client.
What widgets does it include?
It ships seven widgets: basic text, a radio group, a checkbox group, a line chart, an e-commerce product list, a QR code, and a data table, each rendered as an interactive element rather than plain text.
Which chat clients can connect to it?
The project names ChatGPT, Claude Web, Claude Code, Gemini CLI, Cursor, VS Code, Mistral AI, Windsurf and Codex as compatible clients, since any of them can register the tunneled MCP endpoint as a server.
Why does it need a tunnel like ngrok?
The server runs locally on your own machine by default, and most hosted chat clients cannot reach a local address directly, so a tunnel gives the local server a public address the client can register and call.
Can the widgets match my own branding?
Partly. The project includes several predefined themes, default, minimal, enterprise, playful and mono, which change the rendered look of the widgets without a custom styling pass on top of them.
Is this the same as a shadcn or Flowbite component library?
No. It uses Flowbite and Tailwind for styling, but it renders widgets inside a chat client's MCP output rather than installing components into a normal web app's own codebase for a browser to load.


