Chapter 02

The Protocol Landscape

A comprehensive survey of emerging standards for agent communication—examining architectural choices, governance models, and enterprise applicability.

The emergence of multiple agent communication protocols reflects both the urgency of the interoperability challenge and the diverse requirements of different deployment contexts. While this diversity might suggest fragmentation, closer examination reveals complementary design choices addressing distinct layers of the agent communication stack.

Major Protocols

🔌

MCP

Anthropic

Model Context Protocol establishes a standardized interface between AI models and external tools, databases, and APIs—a universal connector for agentic systems.

Agent ↔ Tool

🤝

A2A

Google → Linux Foundation

Agent-to-Agent Protocol enables autonomous agents to discover, authenticate, and collaborate. Supports async workflows, streaming, and webhooks.

Agent ↔ Agent

🔄

ACP

IBM → Linux Foundation

Agent Communication Protocol provides a RESTful wire format for agent tasks. Designed for enterprise REST-centric architectures with token-based auth.

Enterprise REST

⚡

SLIM

Cisco (Outshift)

Secure Low-Latency Interactive Messaging uses gRPC over HTTP/2. Part of Cisco's "Internet of Agents" initiative with streaming and pub/sub patterns.

Low-Latency

🔗

ANP

Emerging Standard

Agent Negotiation Protocol uses JSON-LD and Decentralized Identifiers (DID) for semantic interoperability across heterogeneous domains.

Semantic Interop

💬

AGORA

Research Initiative

Natural language protocol negotiation where agents dynamically agree on communication formats through LLM-mediated generation.

NL Negotiation

Architectural Layering

Understanding the protocol landscape requires recognizing that these standards operate at different architectural layers. MCP addresses vertical integration between agents and external capabilities, while A2A handles horizontal coordination between peer agents.

A2A: Agent Coordination Layer

Task delegation, result sharing, multi-agent workflows

MCP: Tool Execution Layer

API calls, database queries, external integrations

An agent may use MCP to invoke tools during its reasoning process, then use A2A to share results with peer agents—both protocols operating in the same workflow at different abstraction levels.

Protocol Comparison

Protocol	Definition	Complexity	Best Use Cases
A2A	Open agent↔agent tasking and discovery	Moderate; tasks, SSE, agent cards	Cross-vendor multi-agent workflows
MCP	LLM↔tools two-way context bridge	Low; simple client/server	Tool/RAG/IDE integrations
ACP	REST wire format for agent tasks	Moderate; tokens, REST, SSE	Enterprise REST orchestration
SLIM	gRPC-based secure messaging	Higher; gateway, planes, policies	Low-latency event-driven systems
ANP	JSON-LD+DID semantic layer	Higher; JSON-LD, DID, negotiation	Cross-domain semantic interop
AGORA	Natural language negotiation	Specialized; LLM-mediated	Adaptive protocol generation

Governance & Stewardship

Protocol adoption depends not only on technical merit but on governance structures that inspire confidence in long-term stability:

MCP remains under Anthropic's stewardship as an open standard with public specification.

A2A and ACP have transitioned to the Linux Foundation, providing vendor-neutral governance and ensuring no single company controls the standard's evolution.

SLIM operates within Cisco's Outshift initiative, partnering with ecosystem players like LangChain for broader adoption.

This diversity of governance models reflects the nascent state of the ecosystem. As protocols mature, consolidation around common standards seems likely—particularly given Linux Foundation's role in unifying A2A and ACP efforts.