08 · agent-communication · 13 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.
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?
13 as of 2026-09-25: 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).
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.