As AI agents spread, several protocols have emerged to standardise how they connect.
Model Context Protocol (MCP)
Connects AI applications to tools, resources and prompts. The most widely adopted protocol for tool integration.
Agent2Agent (A2A)
Connects agents to other agents for delegation and collaboration, with discovery via Agent Cards and task lifecycles.
Other Efforts
Other proposals address agent communication, agent-driven commerce and payments, and interaction between agents and user interfaces. Some have merged or aligned with others; the landscape is still settling.
How They Layer
- Agent to tool → MCP.
- Agent to agent → A2A.
- Agent to user interface → emerging standards and framework-specific approaches.
- Underneath: HTTP, JSON-RPC, OAuth and standard web infrastructure.
Choosing
Adopt protocols with broad support, open governance and active development. Avoid building on proposals with little adoption unless you have a specific reason.
Stay Current
Specifications change quickly. Track versions, test compatibility and design your systems so protocol layers can be updated without rewriting agent logic.