Human-inspired framework for agent communication

Beyond Message Passing: A Semantic View of Agent Communication Protocols

This paper reframes agent communication through a three-layer lens—communication, syntactic, and semantic—and shows why reliable transport and structured messages are not enough for robust multi-agent systems.

Dun Yuan, Fuyuan Lyu, Ye Yuan, Weixu Zhang, Bowei He, Jiayi Geng, Linfeng Du, Zipeng Sun, Yankai Chen, Changjiang Han, Jikun Kang, Alex Chen, Haolun Wu, Xue Liu
Paper PDF arXiv / OpenReview BibTeX
Core framework
Agent communication and human communication comparison across communication, syntactic, and semantic layers
Figure 1. Agent communication and human communication can be viewed through the same three-layer lens: communication, syntactic structure, and semantic alignment.
Communication Can agents reliably exchange signals, streams, and artifacts?
Syntactic Are tools, capabilities, lifecycle states, and errors represented in a shared structure?
Semantic Do agents align on intent, context, clarification, and verification rather than only payload shape?

Overview

Agent communication protocols are quickly becoming core infrastructure for LLM systems that call tools, collaborate with other agents, and operate across heterogeneous environments. This paper argues that the ecosystem has advanced rapidly on transport and structure, but still lacks strong protocol-level support for meaning alignment. It also provides a practical view of when different protocols are the right deployment choice.

The central claim of the paper is simple: agents should not only exchange messages correctly—they should also have protocol support for understanding them consistently.

Go-to protocol decision map showing when to use agent.json, MCP, ACP-AGNTCY, ACP-IBM, A2C, ANP, LMOS, Agora, WAP, AITP, Coral, and LOKA
Figure 2. Deployment-oriented protocol map summarizing when different protocols are a good fit, including single-tenant use, intra-organizational coordination, trust boundaries, repeated exchanges, web demos, governance, and identity semantics.

What the paper contributes

1. A new lens

The paper introduces a human-inspired framework that maps agent communication onto communication, syntactic, and semantic layers.

2. Comparative analysis

It comparatively analyzes 18 representative protocols spanning tool-use interfaces, agent orchestration standards, decentralized agent networks, and trust-aware systems.

3. A practical diagnosis

It identifies a recurring semantic gap: many protocols standardize delivery and structure, but leave clarification, context alignment, and verification to prompts or wrappers.

Why the semantic layer matters

What current protocols do well

  • Reliable transport
  • Streaming and session mechanics
  • Schema definition and validation
  • Lifecycle and error handling

What often remains missing

  • Clarification when intent is ambiguous
  • Shared contextual grounding across agents
  • Verification that interpreted meaning matches intended meaning
  • Protocol-native repair instead of ad hoc prompting

Protocols analyzed in the paper

The paper spans a wide slice of the emerging agent communication ecosystem, from single-tenant tool protocols to cross-organizational and decentralized designs.

MCP agents.json Agent Protocol ACP-AGNTCY ACP-IBM ACP-AgentUnion ACN A2A Agora LOKA ANP Coral AITP LMOS CrowdES SPPs PXP WAP

Practical angle

Deployment-oriented guidance

The paper also turns its analysis into a deployment-oriented selection workflow, helping practitioners choose a protocol based on settings such as single-tenant tool use, intra-organizational orchestration, open-web delegation, repeated interactions, trust-boundary transactions, or governance requirements.

Design message

The recommendation is to choose the simplest protocol whose native guarantees reduce the relevant communication, schema, and meaning drift—without adding unnecessary complexity.

Abstract

This paper presents a human-inspired perspective on agent communication protocols by organizing them into three layers: communication, syntactic, and semantic. Across 18 representative protocols, it finds a strong ecosystem-wide imbalance: protocol support for transport, streaming, schemas, and lifecycle management has become increasingly mature, while support for clarification, context alignment, and verification remains limited. As a result, semantic responsibilities are often displaced into prompts, wrappers, and application-specific orchestration logic. The paper argues that this creates hidden interoperability and maintenance costs, frames the problem as a semantic gap in the modern protocol stack, and concludes with practical protocol-selection guidance and a broader research agenda for semantically robust agent ecosystems.

Citation

@article{yuan2026beyondmessagepassing,
  title   = {Beyond Message Passing: A Semantic View of Agent Communication Protocols},
  author  = {Yuan, Dun and Lyu, Fuyuan and Yuan, Ye and Zhang, Weixu and He, Bowei and Geng, Jiayi and Du, Linfeng and Sun, Zipeng and Chen, Yankai and Han, Changjiang and Kang, Jikun and Chen, Alex and Wu, Haolun and Liu, Xue},
  year    = {2026},
  journal = {Preprint}
}