Accepted at NeurIPS 2025 AI4NextG Workshop

LLM Agent Communication Protocol (LACP)

Requires Urgent Standardization: A Telecom-Inspired Protocol is Necessary

Authors: Xin Li, Mengbing Liu, Chau Yuen
Nanyang Technological University, Singapore

A unified, telecom-inspired communication protocol ensuring safety, interoperability,
and scalability for LLM agents in Next Generation (NextG) networks.

3.5% Latency Overhead
30% Size Overhead
3 Layer Architecture
Read Paper View Architecture

Abstract

This position paper argues that the field of LLM agents requires a unified, telecom-inspired communication protocol to ensure safety, interoperability, and scalability, especially within the context of Next Generation (NextG) networks. Current ad-hoc communication methods are creating a fragmented ecosystem, reminiscent of the early "protocol wars" in networking, which stifles innovation and poses significant risks.

Drawing inspiration from the layered, standardized protocols that underpin modern telecommunications, we propose the LLM-Agent Communication Protocol (LACP). LACP establishes a three-layer architecture designed to ensure semantic clarity in communication, transactional integrity for complex tasks, and robust, built-in security. Such a standard is critical for ensuring that multi-agent systems can operate safely and reliably in the complex, real-time applications envisioned for 6G and beyond.

The Problem: Fragmented Communication Landscape

The current ecosystem of LLM agent communication resembles the early "protocol wars" of networking, creating a fragmented landscape that stifles innovation and poses significant risks.

Crippling Interoperability Gaps

The absence of a universal standard necessitates bespoke, often brittle, integrations between different agent systems, impeding scalable multi-agent development.

Security as an Afterthought

Security is often not a core component of existing protocols, exposing systems to data tampering, agent spoofing, and adversarial attacks.

Monolithic Design & Lack of Transactional Integrity

Current approaches tightly couple communication logic with agent implementation, leading to systems that are challenging to maintain and extend.

Current Protocol Landscape

OpenAI Function Calling JSON schema, API key auth only
LangChain Agent Protocol REST API, HTTP/JWT auth only
Anthropic MCP JSON-RPC/HTTP, OAuth 2.1
IBM/LF ACP JSON-RPC, Signed capability tokens
Google A2A HTTP/Protobuf, Capability discovery

Key Contributions

Risk Analysis

Identification of systemic risks arising from the fragmented state of current agent communication.

Design Principles

Core guidelines for robust, interoperable agent protocols, distilled from telecommunications history.

Protocol Proposal

LACP, a layered, secure, and extensible framework for LLM-agent interoperability.

Three-Layer Architecture

LACP implements three mutually-insulated layers with well-defined interfaces that enable independent evolution while ensuring system-wide coherence:

1

Semantic Layer

Conveys the intent of communication with universal message types that can be extended with domain-specific content:

PLAN Express high-level intent (intent_id, role, natural_language)
ACT Invoke external tools (intent_id, tool_call, params)
OBSERVE Return results/status (intent_id, status, output)
2

Transactional Layer

Ensures reliability and integrity through comprehensive transaction management:

  • Message signing with JSON Web Signatures (JWS)
  • Message sequencing for ordered delivery
  • Unique transaction IDs for idempotency
  • Two-phase commit for atomic operations
3

Transport Layer

Handles efficient and secure delivery, operating transport-agnostically:

  • HTTP/2 for modern web compatibility
  • QUIC for low-latency applications
  • WebSockets for real-time communication
  • Binary encoding for efficiency
LACP Three-Layer Architecture

Experimental Validation

We implemented a working LACP prototype and conducted comprehensive experiments validating performance, interoperability, and security guarantees.

Performance Analysis

3.5% Latency overhead for complex messages
30% Size overhead for realistic payloads
10,000 Messages tested across payload sizes

LACP's overhead is inversely proportional to message complexity, with minimal impact on realistic agent interactions.

Interoperability Demo

LangChain Agent
→
LACP ACT Message
→
Framework-Agnostic Tool

Successful demonstration of seamless communication between different agent frameworks without custom integration code.

Security Validation

Tampering Attack Blocked by signature verification
Replay Attack Prevented by transaction ID tracking

LACP provides essential application-layer security guarantees beyond what TLS alone can offer.

Insights from Telecommunications

Drawing from the success of telecommunications standardization, LACP applies proven principles that transformed disparate, proprietary systems into globally unified networks.

1G

Channelised Voice

First common control channels (AMPS, NMT) with simple signaling grammar

2G

Digital Identity & Security

GSM introduced digital encryption, SIM cards, and SMS messaging

3G

State Machines for Soft Handover

UMTS/WCDMA enabled packet-switched data and complex state management

4G

All-IP Core

LTE unified voice, video, and data on an all-IP core architecture

5G

Service-Based Architecture

Microservices exposed via HTTP/2 and JSON with network slicing

Key Principles Applied to LACP

Consensus-Driven Standards

Open collaboration prevents vendor lock-in and ensures interoperability

Security by Construction

Security is fundamental, not an add-on, at every protocol layer

Layered Abstractions

Separation of concerns allows independent evolution of system components

Narrow Waist Design

Minimal core ensures stability while extensible edge enables innovation

Implementation & Code

Experimental Validation

We implemented a working LACP prototype and conducted comprehensive experiments validating:

  • Performance overhead analysis (3.5% latency increase for complex messages)
  • Interoperability demonstration with LangChain agents
  • Security validation against tampering and replay attacks
Code Coming Soon Workshop Website

Citation

@inproceedings{li2025lacp,
    title={{LLM} Agent Communication Protocol ({LACP}) Requires Urgent Standardization: A Telecom-Inspired Protocol is Necessary},
    author={Xin Li and Mengbing Liu and Chau Yuen},
    booktitle={NeurIPS 2025 Workshop: AI and ML for Next-Generation Wireless Communications and Networking},
    year={2025},
    url={https://openreview.net/forum?id=LpwE9cSLkS}
}