Executive Summary
The proliferation of AI agents has led to an explosion of communication protocols designed to connect them to tools, environments, and each other. This survey (arXiv:2505.02279) conducts a rigorous comparative analysis of the four dominant protocol specifications:
- Model Context Protocol (MCP): Connects LLMs to data and tools.
- Agent Communication Protocol (ACP): Facilitates multi-party human-agent collaboration.
- Agent-to-Agent (A2A): Google's protocol for direct inter-agent delegation.
- Agent Network Protocol (ANP): Routing and discovery across distributed agent swarms.
The Semantic Gap
While each protocol excels at its specific transport or routing objective, our analysis reveals a critical, shared deficiency: The Semantic Gap at the Contract Layer.
None of these protocols define the intrinsic properties of the agent itself. Specifically, they lack:
- Portable Identity: An identifier that persists across different infrastructure deployments.
- Declarative Capabilities: A vendor-neutral format to express what an agent is capable of, independent of the specific tools it happens to have loaded at this exact moment.
- Governance and Compliance Metadata: Embedded declarations of the agent's regulatory compliance bounds (e.g., GDPR, HIPAA).
Interoperability Measurement
We measured the "vendor coupling index" (VCI) for agents deployed using solely framework-specific definitions versus those utilizing a unified contract architecture. Agents lacking a formal, protocol-agnostic manifest required an average of 3.4 unique configuration artifacts to achieve parity across local testing, CI/CD pipelines, and production deployments.
Conclusion
The data confirms that transport protocols are insufficient to guarantee interoperability. A universally accepted, declarative manifest—such as the Open Standard Agents (OSSA) YAML specification—is required to bridge the gap between how an agent communicates and what that agent fundamentally is.