This paper presents an architectural mediation approach using the Model Context Protocol (MCP) to enable controlled interaction between large language model (LLM) agents and data space services, allowing for interoperable and standards-aligned integration of AI agents into data space ecosystems.