MC
Official✓ VerifiedDeveloper Tools

MCP Server Dev

Design and build MCP servers for Claude. Learn deployment (HTTP, MCPB, local), tools, auth, and interactive apps

Overview

Design and build MCP servers for Claude. Learn deployment (HTTP, MCPB, local), tools, auth, and interactive apps.

A comprehensive guide for designing and building MCP (Model Context Protocol) servers that work seamlessly with Claude. This plugin provides three composable skills that walk you through the entire process — from choosing the right deployment model and tool design patterns to scaffolding code, adding authentication, and packaging for distribution. It covers remote streamable-HTTP servers, interactive MCP apps with embedded UI widgets, and MCPB bundles for local machine access.

The core build-mcp-server skill follows a structured five-phase workflow: it interrogates your use case, recommends a deployment model (remote HTTP, elicitation, MCP app, MCPB, or local stdio), selects a tool-design pattern (one-tool-per-action vs. search+execute for large APIs), picks a framework (TypeScript SDK or FastMCP for Python), and scaffolds your project. The build-mcp-app skill adds interactive UI capabilities — form widgets, searchable pickers, confirmation dialogs, charts, and live-updating status displays — rendered directly in chat via sandboxed iframes. The build-mcpb skill handles packaging local stdio servers with their runtime into installable bundles that require no Node.js or Python on the user’s machine.

How to use: Invoke the entry skill with /mcp-server-dev:build-mcp-server or simply ask “help me build an MCP server” to start the guided workflow. Use /mcp-server-dev:build-mcp-app when you need to add interactive widgets like contact pickers or confirmation dialogs to your server. Use /mcp-server-dev:build-mcpb to package a local server into a distributable MCPB bundle. Each skill includes reference files covering auth flows, widget templates, iframe sandbox constraints, manifest schemas, and security hardening guidelines.

Craft production-grade frontends with distinctive design. Generates polished code that avoids generic AI aesthetics.