Protocols
Four pieces, four different connections.
What each protocol connects, where it stands today, and when a small business needs it. Status checked against the primary sources on 2026-09-25.
A2A (Agent2Agent)
How one agent finds another and asks it to do something.
Connects: Agent to agent, across organizations, over HTTP.
Status: Released specification, version 1.0.0 as of 2026-09-25.
You need it when: You want other people's agents to be able to ask your business a question, request a quote, or start a booking, without a person on either side typing into a form.
A2A Protocol specification (latest) · A2A: agent discovery · A2A project repository
MCP (Model Context Protocol)
How an AI application reaches tools and data.
Connects: An AI application to a server that exposes tools, resources, and prompts.
Status: Published specification, maintained openly.
You need it when: You want an assistant your team already uses to work with your own systems, such as looking up a booking or creating an invoice, through a server you control.
Model Context Protocol specification · Model Context Protocol: introduction
WebMCP
How a web page offers its own functions to an agent in the visitor's browser.
Connects: A page's JavaScript to an agent running in the browser, with the visitor present.
Status: Community Group draft; browser support is early.
You need it when: You want an agent helping a visitor on your website to select a service or request a booking through your page's own functions, instead of guessing at buttons.
llms.txt
A plain-text index of what your site contains.
Connects: A site's pages to any language model or agent that reads the file.
Status: A proposal, widely adopted as a convention; not a formal standard.
You need it when: Always. It is a small file, it costs nothing, and it is the easiest way to tell an agent what is on your site and where your agent card lives.