mcp-builder

verified	true
lastVerifiedAt	"2026-03-15T00:00:00.000Z"
name	mcp-builder
description	Guide developers in creating Model Context Protocol (MCP) servers. Use for building MCP tools that enable LLMs to interact with external services. Covers TypeScript (primary) and Python FastMCP (secondary), tool annotations, Zod/Pydantic validation, and evaluation question creation.
version	1.2.0
model	sonnet
invoked_by	both
user_invocable	true
tools	["Read","Write","Edit","Bash","Glob","Grep","WebFetch"]
agents	["developer","typescript-pro","mcp-developer"]
category	development
tags	["mcp","model-context-protocol","typescript","python","api-integration","tools"]
aliases	["mcp-server-builder","mcp-development"]
best_practices	["Study MCP protocol docs before implementing any server","TypeScript is strongly preferred for its high-quality SDK support","Use streamable HTTP for remote servers, stdio for local servers","Validate all inputs with Zod (TypeScript) or Pydantic (Python)","Annotate tools with readOnlyHint, destructiveHint, idempotentHint, openWorldHint","Create 10 independent evaluation questions per server"]
error_handling	strict
streaming	supported
source	builtin
trust_score	100
provenance_sha	45d11b9bdbe993f2

MCP Server Development Guide

"The quality of an MCP server is measured by how well it enables LLMs to accomplish real-world tasks."

Overview

This skill guides you through creating Model Context Protocol (MCP) servers — tools that enable LLMs to interact with external services. MCP servers expose capabilities as typed tools that agents can discover and invoke.

When to Invoke

• User asks to "build an MCP server" or "create MCP tools"
• Integrating an external API with Claude
• Exposing local system capabilities (files, databases, services) to an agent
• Creating tools for an agent to call

Four-Phase Development Workflow

Phase 1: Deep Research and Planning

Before writing any code, understand the target API or service thoroughly.

1. Study MCP protocol documentation

   • Visit https://modelcontextprotocol.io/sitemap.xml for current docs
   • Understand tool definitions, resource types, and prompt patterns

2. Analyze the target service

   • What operations does the API support?
   • Which operations are read-only vs destructive?
   • What are the authentication requirements?
   • What are the rate limits and pagination patterns?

3. Design tool coverage

   • Balance: API coverage (breadth) vs workflow tools (depth, combining multiple API calls)
   • Prefer workflow tools for common agent tasks
   • Design tool names as snake_case verb-noun pairs: get_user, create_issue, list_repos

4. Choose transport

   • Remote servers: Streamable HTTP transport
   • Local servers: stdio transport
   • Default to stdio unless there's a specific need for HTTP

Phase 2: Implementation

Language Selection

TypeScript (recommended) — high-quality SDK, strong type safety, better IDE support:

import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
import {  }  ;
 { z }  ;

 server =  ({
  : ,
  : ,
});