# The Cost of Saying Hello: An Empirical Study of Handshake Overhead in AI Agent Communication Protocols > Source: > Published: 2026-10-23 07:40:20.024716+00:00 [arXiv](https://arxiv.org/abs/2610.30151) ### The Cost of Saying Hello: An Empirical Study of Handshake Overhead in AI Agent Communication Protocols Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated. Connecting LLM agents via standard communication protocols consumes up to 11% of a 200k-token context window just for the handshake if the counterpart has 200 capabilities, while legacy MCP adds 230ms of latency over three round trips. For production systems shipping multi-agent or tool-heavy architectures, this overhead will rapidly bloat your API bills and degrade user response times. To prevent this, you must force session pinning to achieve zero-RTT warm reconnects or migrate to leaner protocols like A2A that run in a single round trip. A2A reduces handshake overhead to one round trip (61 ms), while legacy MCP requires three round trips (230 ms), making A2A significantly faster for agent communication setups. This directly impacts production latency and context window efficiency, as slower protocols consume up to 11% of a 200k-token context window just on metadata exchange before task execution begins. Optimizing handshake protocols like MCP 2026 or leveraging A2A can drastically reduce latency and token waste, enabling more efficient agent interactions and better utilization of computational resources.