mcp-builder

Guides building MCP servers in TypeScript or Python with evaluation tooling.

Updated Jul 7, 2026
One-click install
npx skills add https://github.com/singhaganesh/Urban-assist --skill mcp-builder-singhaganesh
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: mcp-builder
Source: https://github.com/singhaganesh/Urban-assist/tree/main/.cursor/skills/mcp-builder
Command: npx skills add https://github.com/singhaganesh/Urban-assist --skill mcp-builder-singhaganesh

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve? Building Model Context Protocol servers that LLMs can actually use effectively requires careful tool design, schema validation, transport selection, and testing, which is easy to get wrong without structured guidance. ## Core Features & Use Cases - Four-Phase Development Workflow: Covers research and planning, implementation, review and testing, and evaluation creation for MCP servers. - Language-Specific Guides: Provides reference documentation for TypeScript (MCP SDK with Zod) and Python (FastMCP with Pydantic) implementations. - Evaluation Harness: Includes Python scripts to run XML-based QA evaluations against MCP servers over stdio, SSE, or streamable HTTP transports. - Use Case: You want to expose the GitHub API to an LLM agent. Follow the guide to design tools with proper naming, annotations, and pagination, then generate 10 evaluation questions and run the harness to measure agent accuracy. ## Quick Start Ask the assistant to help you build an MCP server for your chosen API, specifying whether you want TypeScript or Python.

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?

Use the MCP TypeScript SDK's McpServer class and register tools with server.registerTool, providing a title, description, Zod inputSchema, and annotations. Build with tsc and test using the MCP Inspector via npx @modelcontextprotocol/inspector.

How do I test an MCP server with evaluations?

Create an XML file with qa_pair elements containing questions and verifiable answers, then run scripts/evaluation.py with your transport type. The harness connects via stdio, SSE, or HTTP and reports accuracy, tool calls, and agent feedback.

Should I use stdio or streamable HTTP for my MCP server?

Use stdio for local integrations and command-line tools where the server runs as a subprocess. Use streamable HTTP for remote servers serving multiple clients, and avoid SSE since it is deprecated in favor of streamable HTTP.

What makes a good MCP tool design?

Use snake_case names with a service prefix like github_create_issue, provide concise descriptions, validate inputs with Zod or Pydantic, support pagination with limit and offset, and return actionable error messages. Add annotations such as readOnlyHint and destructiveHint.

What are the requirements for MCP evaluation questions?

Questions must be independent, read-only, complex enough to require multiple tool calls, realistic, and have a single stable answer verifiable by string comparison. Avoid questions whose answers change over time, such as counts of current open items.