The interoperability layer for AI agents, explained.
AgentProtocol.ai is a practical guide to AI agent communication standards — MCP, A2A, Agent Protocol and agent interoperability. We track the protocols so you don't have to.
// MCP · A2A · Agent Protocol · interoperability · agent APIs
Three protocols, three different jobs
These are the standards most teams evaluate first. Each solves a distinct part of the agent-communication problem — they often complement rather than compete.
Model Context Protocol
Growing adoptionAn open protocol that standardises how AI applications connect agents and LLMs to external tools, data and context.
- Layer
- Agent ↔ tools/data
- Transport
- JSON-RPC 2.0
- Primitives
- Tools, resources, prompts
Agent2Agent
EmergingAn open protocol for communication and task delegation between independent AI agents, potentially built by different vendors.
- Layer
- Agent ↔ agent
- Transport
- HTTP + JSON-RPC / SSE
- Primitives
- Agent cards, tasks, messages
Agent Protocol
Community specA framework-agnostic REST API specification for interacting with autonomous agents in a consistent, API-style way.
- Layer
- Client ↔ agent (API)
- Transport
- REST / HTTP + JSON
- Primitives
- Runs, tasks, steps, artifacts
What are AI agent protocols?
An AI agent protocol is a shared set of rules for how AI agents communicate — with the tools and data they use, with each other, and with the applications that drive them. Protocols replace one-off, custom integrations with a common language.
Without a protocol, every agent-to-tool or agent-to-agent connection is bespoke. With one, any compliant component can talk to any other — the same idea that let HTTP and REST turn isolated servers into the web.
The space is young and moving quickly. Where standards are still evolving, we say so plainly rather than pretend a single winner exists.
Describe a tool or capability once; any compliant agent can use it.
Components from different vendors cooperate through the shared spec.
MCP, A2A and Agent Protocol operate at different layers and can be combined.
Three layers of agent communication
Most confusion about agent protocols disappears once you separate the layers. Here's where each standard sits.
A protocol like MCP lets an agent discover and call tools, read resources and pull in context — the same way regardless of which tool it is.
A protocol like A2A lets independent agents advertise capabilities, delegate tasks and exchange messages, even across vendors.
An API-style spec like Agent Protocol gives clients a consistent way to start runs, track steps and fetch results from an agent.