mcp-builder

Build MCP servers for Python (FastMCP) or Node/TypeScript (MCP SDK).

146|31|Updated Jul 25, 2025
One-click install
npx skills add https://github.com/bobmatnyc/claude-mpm --skill mcp-builder-bobmatnyc
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: mcp-builder
Source: https://github.com/bobmatnyc/claude-mpm/tree/main/src/claude_mpm/skills/bundled/main/mcp-builder
Command: npx skills add https://github.com/bobmatnyc/claude-mpm --skill mcp-builder-bobmatnyc

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve?

Create high-quality MCP servers that enable LLMs to interact with external services. This guide emphasizes designing tools for agent workflows, not just API wrappers, and optimizing context usage.

Core Features & Use Cases

  • Agent-centric tool design focused on complete workflows
  • Deterministic patterns for Python (FastMCP) and Node/TypeScript (MCP SDK)
  • Evaluation-driven iteration to improve agent usability
  • Context-efficient tool design with actionable error messages

Quick Start

  1. Research protocol docs and MCP docs
  2. Plan agent-centric tools
  3. Implement with validation
  4. Create evaluations
  5. Iterate based on agent feedback

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 that works with LLMs and external APIs?

Build MCP servers using FastMCP for Python or the MCP SDK for Node/TypeScript. Design agent-centric tools with input validation (Pydantic or Zod), proper tool registration, error handling, and support for multiple transports (stdio, SSE, HTTP) to enable LLMs to reliably interact with external services.

What's the best way to design tools for agent workflows in MCP servers?

Design tools for complete workflows rather than simple API wrappers. Use agent-centric patterns with actionable error messages, character-limit awareness, pagination support, and clear tool naming conventions. Validate inputs rigorously and iterate based on agent evaluation feedback to optimize context efficiency.

Can I use FastMCP and the MCP SDK together for different parts of my server?

Yes. FastMCP handles Python implementations while the MCP SDK covers Node/TypeScript. Choose based on your preferred language and existing infrastructure. Both support the same MCP protocol with identical transport options, validation patterns, and tool registration approaches.

What validation tools should I use when building MCP server tools?

Use Pydantic for Python (FastMCP) or Zod for Node/TypeScript (MCP SDK) to validate tool inputs. Both enforce type safety and generate clear error messages that help agents understand and recover from invalid requests, improving workflow reliability.

How do I test if my MCP server tools work well with LLM agents?

Create evaluations to measure how agents interact with your tools. Run agents through representative workflows, collect feedback on tool usability, and iterate on tool design—including naming, descriptions, error handling, and output formatting—based on agent performance metrics.

What transport options does MCP support for connecting servers to agents?

MCP supports stdio, SSE (Server-Sent Events), and HTTP transports. Choose based on your deployment model: stdio for local integrations, SSE or HTTP for remote or cloud-based agent interactions requiring more flexible connectivity.