The AI agent ecosystem is currently in a proliferation phase, with four significant protocols published in the past eighteen months: Model Context Protocol (MCP), Agent Communication Protocol (ACP), Agent2Agent (A2A), and Agent Network Protocol (ANP).
MCP has already won the tool-calling layer, with over 10,000 active public servers and 164 million monthly Python SDK downloads. A2A is a task coordination interface that defines how two agents delegate a task, and has been adopted broadly by enterprise AI teams.
The transport layer remains a problem, with current protocols running over HTTP, which assumes a reachable server and can add latency, cost, and failure modes. New technologies, such as UDP hole-punching and QUIC, are being explored to solve this issue.
