KI wird dann wertvoll, wenn sie über Demos hinausgeht und mit realen Systemen verbunden ist. In diesem Bereich schreibe ich über die Engineering-Seite von Applied AI: MCP, Agenten-Workflows, Tool-Integration, Orchestrierung und die Betriebsanforderungen, die solche Systeme sicher, beobachtbar und in der Praxis nutzbar machen.
Fokusbereiche
Beiträge
MCP Apps: Interactive UIs for AI Clients - A practical guide
With MCP Apps you can add interactive UIs directly into AI clients like Claude and ChatGPT. In this post, I show how to build the Python backend for tools and data, and a React frontend for the interactive UI.
MCP Resources: Adding Climate Data to Your Weather Server
Extend MCP servers with Resources for read-only data. Add 30-year climate normals to the weather server, understand Resource Templates, and combine Tools + Resources for powerful AI workflows.
Agent Skills vs MCP: Two Standards, Two Security Models
A technical analysis comparing Anthropic's Agent Skills specification and Model Context Protocol - two different approaches to extending AI agents, with fundamentally different security implications for credential management and code execution.
Building Your First Agent Skill: A Practical Guide
A hands-on tutorial for creating and using Agent Skills with Claude. Learn how to build a custom weather forecast skill from scratch, complete with Python scripts and API integration.
Publish Once or Fragment Forever? About the State of MCP Registries
From modelcontextprotocol.io's Registry Preview and Generic Registry API to GitHub, Docker, Azure, Smithery, Raycast, MCP.so, PulseMCP, and FluidMCP—what MCP registries actually do and how to choose.
Building Your First MCP Server: A Weather Forecast Integration
A hands-on tutorial for creating production-ready MCP servers with Python. Learn how to build a weather forecast server from scratch, test it locally, containerize it with Docker, and understand the architectural differences from Agent Skills.