AI Agents in Action Foundations for Evaluation and Governance 2025

The landscape of advanced LLM-based agents is supported by new protocols that enable more seamless integration and collaboration. The MCP , for example, aims to standardize the connection between enterprise software systems, external data sources and agents, while protocols such as A2A and the AGNTCY architecture’s agent connect protocol (ACP) offer tools to facilitate interaction between varying AI agents, forming the interoperability layer for multi-agent systems (MAS). As these protocols are implemented across cloud platforms, enterprise networks and edge devices, they are necessary for running agentic code while connecting with real-world sensor data and systems. Introduced by Anthropic in late 2024, MCP6 enables agents to connect with internal or external data sources, APIs and enterprise systems through a standardized protocol. Rather than developing bespoke integrations for each agent-task pairing, MCP allows agents to act as clients that request access to services via MCP-compliant servers. For example, an agent using MCP can check a calendar, retrieve emails, update database content or update CRM records through a shared interface. This significantly reduces friction, speeds up deployment and supports modular plug-and-play capabilities across tools and environments. MCP has gained widespread support across leading agent frameworks and is increasingly viewed as a core mechanism for connecting agents to the broader enterprise infrastructure.1.2 Communication protocols and interoperability Illustration of MCP-based agent communication FIGURE 3 AI agent 1 AI agent 2 User updates a recordOverview of MCP MCP client database MessagingMCP client messaging Send an email MCP server messaging Acknowledge updateMCP server database DatabaseMCP client database Read a record1 MCP server updates the database2 Database update confirmed34 MCP has gained widespread support across leading agent frameworks and is increasingly viewed as a core mechanism. AI Agents in Action: Foundations for Evaluation and Governance 9