mcp-builder

Guides building and evaluating MCP servers in TypeScript or Python.

Updated May 24, 2026
One-click install
npx skills add https://github.com/MWest2020/skill-forge --skill mcp-builder-mwest2020
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: mcp-builder
Source: https://github.com/MWest2020/skill-forge/tree/main/skills/mcp-builder
Command: npx skills add https://github.com/MWest2020/skill-forge --skill mcp-builder-mwest2020

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires anthropic, mcp, and includes scripts (resource) components.

What problem does it solve? Building MCP (Model Context Protocol) servers that LLMs can actually use well requires careful tool design, protocol knowledge, and testing — this Skill provides a structured four-phase process covering research, implementation, review, and evaluation. ## Core Features & Use Cases - Four-Phase Development Workflow: Deep research of the MCP specification and SDKs, implementation with proper schemas and annotations, code review and testing, and evaluation creation. - Language-Specific Guidance: Reference guides for both TypeScript (MCP SDK with Zod) and Python (FastMCP with Pydantic), covering project structure, tool registration, and quality checklists. - Evaluation Harness: Bundled Python scripts connect to MCP servers over stdio, SSE, or streamable HTTP and run XML-defined QA pairs against them using Claude, producing accuracy reports with tool-call metrics. - Use Case: You want to expose a third-party API to Claude as MCP tools. Follow the workflow to design tool names and schemas, implement the server, then run the evaluation script with 10 realistic questions to verify agents can actually use it. ## Quick Start Ask the agent to help you design and implement an MCP server for your chosen API, following the four-phase workflow and loading the TypeScript or Python reference guide.

Frequently Asked Questions about mcp-builder

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I build an MCP server in TypeScript or Python?

Follow the four-phase workflow: research the MCP specification and SDK docs, implement tools with Zod (TypeScript) or Pydantic (Python) schemas, review and test with the MCP Inspector, then create evaluations. Language-specific reference guides cover project structure and complete working examples.

How do I test whether an MCP server works well with LLMs?

Create an XML evaluation file with about 10 complex, read-only, verifiable questions, then run the bundled evaluation.py script against your server. It connects via stdio, SSE, or HTTP, runs each question through Claude with your tools, and reports accuracy, durations, and tool-call metrics.

Should I use streamable HTTP or stdio transport for MCP?

Use streamable HTTP with stateless JSON for remote servers since it scales and is simpler to maintain than stateful sessions. Use stdio for local servers. The evaluation script supports both, plus SSE.

What makes a good MCP tool design for LLM agents?

Use clear action-oriented names with consistent prefixes, concise descriptions, constrained input schemas with examples, output schemas with structuredContent, and actionable error messages. Annotate tools with readOnlyHint, destructiveHint, idempotentHint, and openWorldHint.

What dependencies does the MCP evaluation script require?

The evaluation harness requires the anthropic Python package (>=0.39.0) and the mcp package (>=1.1.0), listed in scripts/requirements.txt. It uses the Anthropic API, so an API key must be configured for the Anthropic client.