Your agent skill is not an anti-corruption layer

2026-06-12 · Source: Thoughtworks Insights · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Advanced, long

Summary

The Model Context Protocol (MCP), often likened to a "USB-C port" for AI integration, is being widely overused as a universal adapter in enterprise AI architectures, according to a June 12, 2026 article. While MCP excels at frictionless adoption for local coding assistants, like the Jira MCP server for VS Code, exposing unmediated upstream schemas directly to an LLM's system prompt introduces significant semantic risk and tight coupling. This "conformist pattern" violates Domain-Driven Design (DDD) principles, leading to architectural fragility, bloated context windows, and increased token costs in production-grade enterprise agent networks. The article advocates for building domain-specific tools as Anti-Corruption Layers (ACLs), leveraging frameworks like Pydantic AI, to define explicit, typed schemas that translate between distinct bounded contexts. This approach enforces ubiquitous language, provides built-in input validation, and uses established protocols like REST or gRPC, ensuring more robust and reliable enterprise AI systems, while still acknowledging MCP's value for rapid experimentation.

Key takeaway

For AI Architects designing enterprise agent networks, avoid using raw Model Context Protocol (MCP) servers as direct integration points. Your agent's system prompt is not an anti-corruption layer; relying on it introduces significant architectural fragility and semantic risk. Instead, build domain-specific tools with explicit, typed schemas to act as Anti-Corruption Layers, ensuring robust data validation and clear boundaries. This approach enhances reliability and reduces token costs for production-grade AI applications.

Key insights

MCP's direct exposure of upstream schemas to LLMs creates architectural debt; use domain-specific tools as Anti-Corruption Layers for robust enterprise AI.

Principles

• Bounded contexts define explicit domain model boundaries.
• Conformist pattern tightly couples systems, risking fragility.
• Anti-corruption layers translate data, preserving integrity.

Method

Build domain-specific tools with explicit, typed schemas (e.g., via Pydantic AI) to act as an Anti-Corruption Layer. This translates between bounded contexts, enforces ubiquitous language, and validates inputs before LLM interaction.

In practice

• Define custom tools matching your agent's ubiquitous language.
• Use robust data validation for tool interfaces.
• Employ established protocols like REST, GraphQL, gRPC.

Topics

• Model Context Protocol
• Domain-Driven Design
• Anti-Corruption Layer
• Enterprise AI Agents
• Bounded Contexts
• Pydantic AI

Code references

• atlassian/atlassian-mcp-server

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, AI Architect, Software Engineer

Related on AIssential

• MCP Is Dead — MCP's design of injecting tool descriptions into an agent's context window creates a critical, unpatchable security vulnerability.
• Claude Code: The 2026 Zero to Hero Guide — Claude Code transforms engineering by enabling autonomous, agentic code development and system orchestration.
• The Agent Stack - Part 6: Tools, MCP, and Capability Surfaces — AI tools are capability surfaces, not just function calls, requiring careful boundary design for security and control.
• Rebooting Enterprise AI with MCP and Kubernetes — MCP acts as a "selectively permeable membrane" enabling secure, controlled AI agent interaction with enterprise systems.
• We Don’t Have an AI Problem. We Have a Context Problem. (And MCP Exposes It) — Unstructured context, not model limitations, is the primary bottleneck in scaling and maintaining LLM-based systems.
• AI Agents, Tools, MCP, and Skills: The Core, The Embellishment, and The Gimmick — AI agent development requires distinguishing core components like Agents and Tools from less practical elements like MCP.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Thoughtworks Insights.