Sales Explorer
A minimal MCP server that demonstrates an interactive bar chart UI for sales data exploration, enabling drill-down by clicking a region without requiring additional model turns.
README
Sales Explorer — a minimal MCP App
A tiny, dependency-light demo of an MCP App: an interactive UI that an MCP server ships into an AI host (Claude, etc.), rendered in a sandboxed iframe right in the conversation.
The whole point is one moment: you click a bar and the chart drills down with no model round-trip — and the assistant still knows what you're looking at.
The three beats this demonstrates
- UI in the chat, not text. Ask "show me sales by region" → a bar chart renders inline instead of a table of numbers.
- Bidirectional, model-free interaction. Click a region → the iframe calls
the
get_salestool itself and re-renders. No new prompt, no model turn. - The model stays in the loop. After drilling in, the app tells the host what you're viewing, so you can then say "email this to my team" and it knows what "this" is.
How it works (the two files that matter)
server.ts— registers a tool (get_sales) whose_meta.ui.resourceUripoints at a UI resource (ui://sales-explorer/mcp-app.html). That one line of metadata is what turns a normal tool into an MCP App.src/mcp-app.ts— the iframe app. It renders the chart from the result the host pushes in (app.ontoolresult), and on a bar click callsapp.callServerTool({ name: "get_sales", arguments: { region } }).
The UI is bundled by Vite into a single self-contained HTML file
(dist/mcp-app.html) so the server can return it as one resource with no
external asset requests.
Run it
Requires Node ≥ 22.18 (run TypeScript directly — no tsx/ts-node).
npm install
npm run dev # builds the UI on change + serves MCP over HTTP on :3001
- HTTP endpoint:
http://localhost:3001/mcp(Streamable HTTP, stateless) - stdio:
npm run start:stdio
Then point an MCP-Apps-capable host at it (Claude Desktop, MCPJam, VS Code Copilot, …) and ask: "show me sales by region."
Verify without a host (curl)
npm start & # build UI + serve
# list tools — note the _meta.ui.resourceUri on get_sales
curl -s -H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-X POST http://localhost:3001/mcp \
--data '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
# the drill-down the iframe performs on a click
curl -s -H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-X POST http://localhost:3001/mcp \
--data '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"get_sales","arguments":{"region":"EMEA"}}}'
Project layout
server.ts # MCP server: get_sales tool + ui:// resource
main.ts # transport (Streamable HTTP / stdio)
mcp-app.html # UI shell (bundled into dist/mcp-app.html)
src/mcp-app.ts # the iframe app: render chart, click -> callServerTool
src/styles.css # chart styling (uses host theme variables when provided)
vite.config.ts # single-file bundle config
Built on @modelcontextprotocol/ext-apps + @modelcontextprotocol/sdk.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。