MCP solved tool calling. A2A solved coordination. What solves transport?

2026-06-14 · Source: VentureBeat · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation · Depth: Intermediate, short

Summary

The AI agent ecosystem is experiencing a proliferation of communication protocols, mirroring historical trends in distributed computing. Four significant protocols emerged recently: Anthropic's Model Context Protocol (MCP) for tool-calling, IBM Research's Agent Communication Protocol (ACP) for lightweight messaging, Google's Agent2Agent (A2A) for task coordination, and the independent Agent Network Protocol (ANP) for discovery and identity. MCP has already won the tool-calling layer with over 10,000 active public servers and 164 million monthly Python SDK downloads by April 2026. While these protocols address different layers of an emerging stack, a critical transport layer problem remains. Current HTTP-based protocols struggle with direct peer-to-peer communication behind Network Address Translation (NAT), affecting 88% of networked devices. Solutions like UDP hole-punching, STUN, QUIC, and capability-based routing are being explored, with Pilot Protocol and libp2p leading efforts. HTTP-based protocols will stabilize within 12 months, but the transport layer is 18-24 months behind, with standardization expected in 2027-2028.

Key takeaway

For engineering leaders making architecture decisions today, you should confidently adopt stable application-layer protocols like MCP for tool-calling. When implementing multi-agent coordination, use A2A but anticipate its continued evolution. Critically, design your agent systems with a clean separation between application semantics and the underlying transport layer. This architectural foresight is inexpensive now and will provide significant flexibility when the peer-to-peer transport layer eventually stabilizes in the next 18-24 months, avoiding costly retrofits.

Key insights

The AI agent ecosystem is converging on a layered protocol stack, but a critical transport layer problem remains.

Principles

• Distributed computing protocols evolve from proliferation to consolidation.
• AI agent protocols often address distinct layers of a communication stack.
• Architectural separation of application semantics and transport is vital.

In practice

• Adopt MCP for agent tool-calling due to its stability.
• Implement A2A for multi-agent task coordination, anticipating evolution.
• Design agent systems with clean application/transport layer separation.

Topics

• AI Agents
• Agent Protocols
• Distributed Systems
• NAT Traversal
• P2P Networking
• MCP
• A2A

Best for: CTO, VP of Engineering/Data, AI Product Manager, AI Architect, AI Engineer, Director of AI/ML

Related on AIssential

• A2A vs MCP: AI Agent Communication Explained — A2A and MCP are complementary protocols enabling AI agent interoperability and standardized external resource access.
• Toward a Modular Architecture for Embedded AI Agent Systems at the Edge — A modular architecture bridges real-time control and agentic AI for embedded systems using tiered agents and a governance layer.
• The Protocol That Cleaned Up Our Agent Architecture — The Model Context Protocol (MCP) centralizes AI agent tool management, decoupling tool definitions from orchestrators for improved interoperability and scalability.
• Securing AI agents: How AWS and Cisco AI Defense scale MCP and A2A deployments — Automated security scanning and unified governance are crucial for scaling enterprise AI agent deployments securely.
• MCP vs Agent Skills: Different Altogether — MCP and Agent Skills are complementary, not competing, technologies for AI agent development.
• What Is the A2A Protocol? — A2A standardizes AI agent collaboration, enabling secure, interoperable communication and task management across platforms.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by VentureBeat.