my-mcp-server
A minimal MCP server for task management with a widget-based UI, built using the @miragon/mcp-toolkit.
README
my-mcp-server — minimal @miragon/mcp-toolkit starter
A self-contained starter for an MCP server built on the published
@miragon/mcp-toolkit packages: one host (createFrameworkApp), one module
that registers its own tools plus a widget (the tasks module), and the
Vite setup that builds the widget bundle into a single mcp-app.html.
Nothing in here depends on the monorepo.
Two ways to start:
- Use the template repo —
Miragon/mcp-toolkit-startermirrors this directory. Click "Use this template" (orgh repo create my-mcp-server --template Miragon/mcp-toolkit-starter) and you have a fresh repo with CI included. - Copy this directory out of the
mcp-toolkitrepo and rename it.
The included CI (.github/workflows/ci.yml) typechecks and builds the widget
bundle. It installs from the restricted @miragon scope: repos in the Miragon
org can use the built-in workflow token once the packages grant them read
access; anywhere else, add a PACKAGES_READ_TOKEN repo secret (a PAT with
read:packages).
Prerequisites
- Node.js 20 or newer, and pnpm.
- A GitHub personal access token with
the
read:packagesscope, exported asGITHUB_TOKEN. The packages live on GitHub Packages under the restricted@miragonscope; the.npmrcin this directory is already wired to it.
export GITHUB_TOKEN=ghp_…
Run it
pnpm install
pnpm start # builds the widget bundle, then boots the host
Then open http://localhost:3010/inspector (the Inspector is built into
mcp-use) and call show_tasks_board. That is the full loop: an MCP tool
returning a rendered widget.
Optional config lives in .env (copy env.example first):
cp env.example .env
Project layout
├── package.json # pinned versions; dev / build:bundle / start / typecheck
├── .npmrc # @miragon scope → GitHub Packages
├── env.example # PORT / MCP_URL — copy to .env
├── src/
│ ├── host.ts # createFrameworkApp: plugins, proxies, app bundle
│ └── modules/tasks/
│ ├── definition.ts # static contract: module name + widget ids
│ ├── tool-names.ts # tool-name constants (server ↔ widget agree here)
│ ├── store.ts # in-memory domain layer; TasksBoardData view-model
│ ├── plugin.ts # tools: list_tasks, create_task, show_tasks_board, tasks_board_data
│ └── widgets/TaskListCard.tsx # the widget: {data: TasksBoardData} → UI primitives
└── app-bundle/
├── index.html # bundle entry + React importmap shim
├── main.tsx # widget-id map → React components (McpToolkitApp)
├── main.css # Tailwind entry: globals.css + @source scan paths
├── vite.config.ts # single-file build → app-bundle/dist/index.html
└── vite-env.d.ts
How the widget bundle works
The host serves one HTML file — mcp-app.html — as an MCP resource
(app.resourceUri in src/host.ts). MCP clients render it in an iframe fed
from the resource content, so there is no second HTTP request for chunks:
the bundle must be a single self-contained file. That is what
app-bundle/vite.config.ts produces (vite-plugin-singlefile, everything
inlined) into app-bundle/dist/index.html, which htmlPath points at.
Inside the bundle, app-bundle/main.tsx maps widget ids to React components:
- Every widget is registered twice: once in the plugin's
definition.ts(the id + theconsumesdataType the server pushes) and once in the bundle's widget map (the id + the component). The two sides meet on the widget id —"tasks:board"here. adaptDataWidget(TaskListCard, "tasks:board")resolves the step whose_dataTypeis"tasks:board"(set bybuildSingleWidgetViewinshow_tasks_board) and forwards its data to the component'sdataprop.
The bundle is a build artifact: after changing a widget or the map, rebuild it
(pnpm build:bundle) and restart — the host reads the file at
app-bundle/dist/index.html. pnpm start chains both.
Styling: app-bundle/main.css imports the toolkit's
@miragon/mcp-toolkit-ui/globals.css (Tailwind theme + tokens) and adds two
@source lines so Tailwind generates classes used outside the CSS file's own
tree — your widgets under src/, and the UI package's shipped sources in
node_modules. If a class "does nothing" in the rendered widget, check that
the file using it is covered by an @source line.
Where to go next
- Docs — concepts, guides, and the API reference for every package.
- The
tasksexample — the full-size version of this module (complete_task, filterable board widget, in-widget refresh, tests). - Agent skills
— the repo's coding-agent skills (
build-mcp-server,add-mcp-tool,build-mcp-widget,compose-a-view,white-label-client) encode the house patterns; copy them into this project's.claude/skills/so your coding agent builds on them.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.