Data Spaces enable sovereign, governed data sharing across organizational boundaries, but their interaction model clashes with large language model (LLM) agents, which operate via probabilistic natural‑language dialogs. To bridge this gap, we propose an architectural mediation layer built on the Model Context Protocol (MCP) and realized through the Eunomia Agent.
Eunomia sits between the LLM and the Data Space, translating Data Space capabilities—catalog lookup, metadata retrieval, data service invocation—into structured, JSON‑Schema‑driven tool definitions that LLMs can discover and invoke while respecting governance constraints. The protocol carries permission tags and compliance checks, guaranteeing that every interaction adheres to the Data Space’s policy model.
Core technical components are:
- Capability abstraction: MCP
Capabilityobjects describe service inputs/outputs via schemas. - Tool generation: Eunomia dynamically emits OpenAI‑compatible function specifications, allowing the LLM to call them directly in a conversation.
- Governance interception: Each call is vetted against permission tags; non‑compliant requests are rejected or rewritten.
A prototype validates end‑to‑end interaction on a standard Data Space instance: the LLM triggers catalog discovery → fetches dataset metadata → issues a constrained data access request, all without modifying existing Data Space components. Results show that protocol‑based mediation provides plug‑and‑play interoperability while preserving standards compliance.
The approach offers organizations a practical blueprint for introducing AI‑driven automation into governed data‑sharing environments: the mediation layer enforces separation of concerns, maintaining data governance integrity while exposing programmable LLM capabilities.
Review: The paper presents a protocol‑centric design that reconciles compliance with flexibility, laying a solid technical foundation for deeper AI and Data Space integration.