08 · agent-communication · 18 systems
Agent communication & collaborationProtocols and frameworks for agents that talk to agents.
Bridges connect agents to tools; this category connects agents to each other. It splits into open protocols for cross-vendor discovery and messaging (A2A, AGNTCY, ANP, NANDA) and multi-agent frameworks where a supervisor delegates work to specialists through handoffs (CrewAI, Microsoft Agent Framework, LangGraph, MetaGPT, CAMEL).
Systems in this category
OpenScout / Scout
OpenScout · agent-communication
OpenScout / Scout is a local-first broker that lets addressable agents (Claude Code, Codex, Cursor) send and ask across harnesses on one mesh.

A2A Protocol
A2A Project · agent-communication
A2A (Agent2Agent) is the open protocol for cross-vendor agent communication, Google-originated, now governed at the Linux Foundation, standardizing Agent Cards for capability discovery plus task-based messaging between agents.
AGNTCY
AGNTCY · agent-communication
AGNTCY is the Cisco-originated 'Internet of Agents' collective, an open-source stack for agent discovery (directory), secure messaging (SLIM), identity, and observability across vendors.

Agent Network Protocol
Community / OSS · agent-communication
Agent Network Protocol (ANP) is an open spec for a decentralized web of agents, DID-based identity, capability description, and peer-to-peer agent messaging without a central directory.

NANDA
NANDA (MIT Media Lab) · agent-communication
NANDA is MIT Media Lab's 'Internet of AI Agents' research program, a registry and index layer plus protocols for discovering, verifying, and composing agents at internet scale.
BeeAI
IBM · agent-communication
BeeAI is IBM's open-source agent platform, run, compose, and observe agents locally or on a cluster, with ACP/A2A-style interop and a web UI for non-technical operators.
CrewAI
CrewAI · agent-communication
CrewAI is the most widely adopted multi-agent framework: model crews of role-based agents in YAML or Python, run sequential or hierarchical processes, and manage fleets through the CrewAI control plane.
Microsoft Agent Framework
Microsoft · agent-communication
Microsoft Agent Framework is the successor to AutoGen and Semantic Kernel's agent pieces, one OSS framework (Python and .NET) for single agents, multi-agent graphs, and Magentic-style orchestration.

LangGraph
LangChain · agent-communication
LangGraph is LangChain's agent orchestration framework, durable stateful graphs with supervisor and swarm multi-agent patterns, checkpointing, and LangGraph Platform for deployment.

MetaGPT
DeepWisdom · agent-communication
MetaGPT is the multi-agent framework that turns a one-line brief into a simulated software company, PM, architect, and engineer agents handing off artifacts along an assembly line.
CAMEL
CAMEL-AI · agent-communication
CAMEL is the research-rooted multi-agent framework: role-playing agents, agent societies, and toolkits for scaling multi-agent behavior, widely used in academic multi-agent studies.

Agency Swarm
VRSEN · agent-communication
Agency Swarm is VRSEN's framework for building 'agencies' of OpenAI-powered agents, defining agents, their tools, and which agents are allowed to talk to which over an explicit communication-flow graph.

OpenAI Agents SDK
OpenAI · agent-communication
The OpenAI Agents SDK is OpenAI's lightweight framework for agentic apps, its handoffs primitive is a first-class agent-to-agent mechanism: one agent transfers the live conversation to a specialist.

Reason Network
Reason Machines · agent-communication
Reason Network connects coding agents across a team through agent discovery and messaging.

Agent Relay
AgentWorkforce · agent-communication
Agent Relay connects coding agents through shared channels, direct messages, and coordinated workflows.

hcom
aannoo · agent-communication
hcom lets coding agents message, observe, and launch peers across terminal sessions.

Claude Peers MCP
louislva · agent-communication
Claude Peers MCP gives Claude Code sessions a way to discover and message one another.

MCP Agent Mail
Dicklesworthstone · agent-communication
MCP Agent Mail provides coding agents with identities, inboxes, searchable threads, and advisory file reservations.
Who it's for
Builders composing several agents into one system — a coordinator plus specialists — and teams that need agents from different vendors or runtimes to discover and message each other.
Tradeoffs
Frameworks are production-ready code with strong opinions; cross-vendor protocols are early and still converging (IBM's ACP folded into A2A). Most shipping 'multi-agent' systems today are single-framework orchestration — true agent-to-agent messaging across organizations is emerging.
Frequently asked
What is an agent communication platform?
Bridges connect agents to tools; this category connects agents to each other. It splits into open protocols for cross-vendor discovery and messaging (A2A, AGNTCY, ANP, NANDA) and multi-agent frameworks where a supervisor delegates work to specialists through handoffs (CrewAI, Microsoft Agent Framework, LangGraph, MetaGPT, CAMEL).
Who should use an agent communication platform?
Builders composing several agents into one system — a coordinator plus specialists — and teams that need agents from different vendors or runtimes to discover and message each other.
What are the tradeoffs of agent communication platforms?
Frameworks are production-ready code with strong opinions; cross-vendor protocols are early and still converging (IBM's ACP folded into A2A). Most shipping 'multi-agent' systems today are single-framework orchestration — true agent-to-agent messaging across organizations is emerging.
Which agent communication platforms are in the catalog?
18 as of 2026-10-19: OpenScout / Scout (OpenScout), A2A Protocol (A2A Project), AGNTCY (AGNTCY), Agent Network Protocol (Community / OSS), NANDA (NANDA (MIT Media Lab)), BeeAI (IBM), CrewAI (CrewAI), Microsoft Agent Framework (Microsoft), LangGraph (LangChain), MetaGPT (DeepWisdom), CAMEL (CAMEL-AI), Agency Swarm (VRSEN), OpenAI Agents SDK (OpenAI), Reason Network (Reason Machines), Agent Relay (AgentWorkforce), hcom (aannoo), Claude Peers MCP (louislva), MCP Agent Mail (Dicklesworthstone).
How is this different from bridges and MCP?
Bridges are agent↔tool: MCP gives one agent access to tools and data. Agent communication is agent↔agent: discovery, messaging, and delegation between autonomous systems — often one agent calling another the way it would call a tool.
Is A2A replacing MCP?
No — different layers. MCP connects an agent to tools; A2A connects agents to agents. The emerging pattern is MCP inside the agent for tool calls and A2A between agents for task handoffs.
Do I need a multi-agent framework to run several agents?
No — a queue, broker, or shared channel is enough for simple fan-out. Frameworks like CrewAI, Agent Framework, or LangGraph add role and goal abstractions, handoff logic, shared state, and tracing — which pays off once specialists outnumber the coordinator.