mcp-builder

Build, configure, and register MCP servers for Claude Code.

Updated Aug 2, 2026
One-click install
npx skills add https://github.com/leonardoacosta/agents --skill mcp-builder-leonardoacosta
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: mcp-builder
Source: https://github.com/leonardoacosta/agents/tree/main/skills/mcp-builder
Command: npx skills add https://github.com/leonardoacosta/agents --skill mcp-builder-leonardoacosta

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve? Deciding whether to build an MCP server, choosing the right transport, and wiring it into Claude Code's mcp.json involves many failure-prone details such as auth token handling, Docker isolation, and tool description quality. This Skill provides decision trees, config patterns, and implementation templates that prevent common mistakes. ## Core Features & Use Cases - Decision Framework: A decision tree for choosing between MCP, Bash, WebFetch, and sub-agents based on auth needs, call frequency, and latency. - Config Patterns: Ready-to-use mcp.json patterns for Docker containers, mcp-remote OAuth services, and local stdio processes with env var injection. - TypeScript Server Template: Guidance for building MCP servers with @modelcontextprotocol/sdk, Zod validation, tool annotations, and pagination. - Evaluation Harness: Python scripts to create and run XML-based evaluations that test whether LLMs can effectively use your MCP server's tools. - Use Case: You need to call an authenticated GitHub-like API repeatedly in a session. Use this Skill to decide MCP is warranted, scaffold a TypeScript server, configure it in mcp.json with Docker isolation, and verify it with claude mcp list. ## Quick Start Ask the agent to help you build and register a new MCP server for an external API in Claude Code, starting with the MCP-versus-Bash decision tree.

Frequently Asked Questions about mcp-builder

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

FAQPage Schema
When should I build an MCP server instead of using Bash or WebFetch?

Build an MCP server when an external service requires persistent auth and is called more than five times per session. Use WebFetch for single public API calls and Bash for local filesystem operations or one-off commands.

How do I configure an MCP server in Claude Code mcp.json?

Add an entry under mcpServers in ~/.claude/mcp.json with a command, args, and env block. Use Docker with -i and --rm for write-access servers, npx mcp-remote for OAuth services, and absolute paths for local node processes.

How do I pass API keys to an MCP server securely?

Use the ${VAR} syntax in the env block of mcp.json, which reads values from the shell environment where Claude Code launched. Never hardcode tokens directly in the config file.

Why does my Docker-based MCP server hang on the first tool call?

The most common cause is a missing -i flag in the docker run args, which closes stdin and breaks stdio transport. Also add --rm to prevent orphaned containers from accumulating.

How do I test whether my MCP server tools work well with LLMs?

Create an XML evaluation file with 10 read-only, complex questions having single verifiable answers, then run scripts/evaluation.py against your server via stdio, SSE, or HTTP transport. The harness reports accuracy, tool call counts, and agent feedback.

What transport should a remote MCP server use?

Use streamable HTTP for remote servers serving multiple clients, and stdio for local single-session integrations. SSE is deprecated in favor of streamable HTTP, though mcp-remote still uses SSE endpoints for OAuth services.