commercetools announced Managed MCP Servers, a hosted approach to deploying Model Context Protocol (MCP) agents that restricts each agent to only the tools and data fields it needs for a specific job (commercetools Blog). Unlike the generic Commerce MCP surface—which exposes over 100 tools and all available data—Managed MCP Servers let teams remove unnecessary tools, redact sensitive fields (such as email addresses or payment details), fix parameters to constant values, and rewrite tool descriptions with project-specific knowledge (commercetools Blog).
The commercetools MCP skill, part of commercetools for Builders, automates the configuration process by analyzing a project's real data, proposing a minimal tool set and field filters, and then testing the server with blind-run agents to iterate on what works (commercetools Blog). For commerce practitioners, this shift moves security enforcement outside the model—so agents cannot bypass restrictions by calling the API directly—while reducing token consumption and keeping personal data away from models that have no reason to see it (commercetools Blog). The approach supports both human-in-the-loop workflows in Claude Desktop and autonomous deployments on platforms like Amazon Bedrock, with observability through Platform Insights to track agent usage in production (commercetools Blog).