mcp-builder

name	mcp-builder
description	MCP server development for AI agents. Designs tool schemas, implements Python/TypeScript servers, creates evaluation tests. Use when user asks to build MCP server, create tool integration, or develop Claude plugins. Supports GitHub/Notion/Slack integrations.

MCP Server Development Tool

Build high-quality Model Context Protocol servers to provide external tool capabilities for AI agents.

Core Capabilities

• MCP protocol understanding and server architecture design
• Python/TypeScript SDK implementation
• Tool schema design for AI agents
• Evaluation test creation and validation

Tech Stack

Language	SDK	Validation	Runtime
Python	MCP Python SDK	Pydantic v2	asyncio
TypeScript	MCP TypeScript SDK	Zod	Node.js 18+

Executable Tools

The following scripts can be run directly without reading source code:

• scripts/evaluation.py - Run MCP server evaluation tests
• scripts/connections.py - Test server connection status

Design Principles

1. Workflow-Oriented: Build complete task tools, not simple API wrappers
2. Optimize Context: Return high-signal information, avoid data dumping
3. Actionable Errors: Error messages guide agents to correct usage
4. Evaluation-Driven: Create evaluation scenarios early, iterate based on agent feedback

Boundaries

Focus on MCP server development and tool design, not third-party API development or client integration.

When NOT to Use

• Writing general business logic code → use developer
• Frontend UI development → use frontend-design
• API design → use api-designer
• Database schema design → use database-engineer
• Product planning or requirements → use product-manager

Detailed References

• ./workflows/mcp-development.md - Complete development workflow
• ./guides/mcp_best_practices.md - MCP best practices
• ./guides/python_mcp_server.md - Python implementation guide
• ./guides/node_mcp_server.md - TypeScript implementation guide
• ./guides/evaluation.md - Evaluation creation guide

Escalation Rules

Pause and ask the owner before:

• exposing tools that have overly broad permissions or weak schema boundaries
• expanding server scope into generic third-party API integration work
• shipping an MCP interface without at least a basic evaluation or validation path

Final Output Contract (MANDATORY)

Every use of this skill should end with:

1. Skill Fit - why an MCP server or tool design solution is needed
2. Primary Deliverable - server design, tool schema, or implementation plan
3. Execution Evidence - guides used, files changed, and evaluation steps prepared
4. Risks / Open Questions - permission, schema, or agent-usage concerns
5. Next Action - the next build, test, or review step