AI/ML Model Cards in Edge AI Cyberinfrastructure: towards Agentic AI

TL;DR

This work treats AI/ML model cards as dynamic, lifecycle-aware objects embedded in the ICICLE edge-AI ecosystem via Patra Model Cards. It evaluates the Model Context Protocol (MCP) as a session-based interface to the Patra server, contrasting its performance with REST and exploring how persistent sessions enable agentic, provenance-aware workflows at the edge. Through microbenchmarks and real-world tests, the study highlights latency tradeoffs, scalability considerations, and the potential value of MCP for continuous model-use benchmarking and automated, context-driven model selection. The findings inform design choices for agent-oriented AI infrastructures and point to future work on ownership signaling, schema coupling, and fuller integration of MCP into edge-aware MLOps pipelines.

Abstract

AI/ML model cards can contain a benchmarked evaluation of an AI/ML model against intended use but a one time assessment during model training does not get at how and where a model is actually used over its lifetime. Through Patra Model Cards embedded in the ICICLE AI Institute software ecosystem we study model cards as dynamic objects. The study reported here assesses the benefits and tradeoffs of adopting the Model Context Protocol (MCP) as an interface to the Patra Model Card server. Quantitative assessment shows the overhead of MCP as compared to a REST interface. The core question however is of active sessions enabled by MCP; this is a qualitative question of fit and use in the context of dynamic model cards that we address as well.

Figures (9)

• Figure 1: AI/ML model object lifecycle
• Figure 2: AI/ML model development-deployment pipeline with model card collection points
• Figure 3: Core Entities of the Patra Model Card
• Figure 4: Inference at the Edge through ML Workbench
• Figure 5: Model card retrieval, few KB model card sizes