Scalable and Observable Automation with MCP

In the era of AI-native Automation, building scalable, interoperable systems is no longer optional, it’s essential. As enterprise users increasingly adopt large language models (LLMs) to drive intelligent operations, the need for a unified, composable protocol becomes critical.

The Model Context Protocol (MCP) is emerging as the de-facto standard for enabling LLMs and AI agents to interact with automation tools. Think of it as the “USB-C” of automation: plug-and-play, secure, and built for modularity.

This blog provides practical guidance for integrating MCP into your products. We explore multiple MCP architecture patterns including Product-Level and Component-Level servers, alongside Remote MCP Server deployment models and observability strategies using OpenTelemetry. Whether you’re designing for internal microservices or exposing APIs to external LLM agents, this post offers actionable insights to help you build reliable, future proof automation experiences with MCP.

MCP Server Levels: Two Common Patterns

Product-Level MCP Server

Definition

A single MCP server that wraps and exposes only the stable, supported, and public-facing APIs of an entire product.

Examples

• Instana MCP Server
• Dynatrace MCP Server
• Grafana MCP Server