Protocols for agents
I keep this page as an index. Each protocol or format sits next to its official page, and physical AI enters only with what those official pages actually describe.
The For agents page explains what an agent can read on this site. This page is the map of the protocols, not a guide to reading this domain.
MCP
The Model Context Protocol, or MCP, is an open protocol. The official specification describes it as a standardized way to connect language model applications to data sources and tools. The spec itself says it takes some inspiration from the Language Server Protocol, which organized language support across development tools. The analogy is about standardizing an ecosystem.
A2A
Agent2Agent, or A2A, is an open standard for communication between independent agent systems, even when they were built in different frameworks or by different vendors. The specification published on the project's site is 1.0.0. A2A itself describes the fit with MCP: MCP connects an agent to tools, APIs and data; A2A connects an agent to other agents. The two are used together.
AGENTS.md
AGENTS.md is an open Markdown format for guiding coding agents, a kind of README for the agent. It is stewarded by the Agentic AI Foundation, or AAIF, under the Linux Foundation, the same umbrella as MCP and A2A. It belongs with project instructions, not with conversation between agents.
Poppy
Personal Agent Protocol, or Poppy, in version 1.0 published by Sierra, describes how personal agents interact with companies through business agents, websites and APIs. In Sierra's account, the protocol is built in large part on existing standards such as MCP, OpenAPI and the Universal Commerce Protocol (UCP). That is the cut of a personal agent arriving at a company.
WebMCP
WebMCP appears in a draft from the Web Machine Learning Community Group. That text describes an API that lets a page offer JavaScript tools to an agent already in the browser. In the draft, the page itself can work as an MCP server, with the tools running in the page's script rather than on a separate server. The document itself states that it is not a W3C Standard and is not on the W3C Standards Track.
ARD
Agentic Resource Discovery, or ARD, is a specification, in proposal status, for cataloging what a domain offers to agents, such as an MCP server. The current path for the manifest is /.well-known/ard.json, announced with rel="ard". The manifest describes resources; the tool itself lives elsewhere, for example on an MCP server.
Physical AI
When the subject turns to physical AI, official sources support another layer, equally real, only of a different kind. NVIDIA defines physical AI as what enables autonomous machines, such as robots in factories and robotaxis, to perceive, understand and perform complex actions in the physical world. What those pages describe clearly are platforms, models, robotics kits and simulation and data standards. There is not yet, in the same sense as MCP or A2A, a shared protocol for agents in the physical world.
New protocols enter this index only when they have an official page.