How They Fit Together

These seven standards aren't competing—they're layers. Each solves a different problem in the AI stack. Together, they create an open ecosystem where agents can use any tool, coordinate with other agents, and interact with users through rich interfaces.

Configuration AGENTS.md
UI Generation Open-JSON-UI
UI Rendering A2UI
Interaction AG-UI
Coordination A2A
Tool Integration MCP · UTCP

See Them Work Together

Follow one request through the full stack.
"Book a conference room for our team standup next Tuesday."

AGENTS.md Before anything runs, the AI assistant reads the project's AGENTS.md file. It learns which agents exist, how to authenticate with them, and what conventions to follow.
A2A The assistant discovers a Facilities agent by fetching its Agent Card. It creates a Task—"book a room for the Tuesday standup"—and delegates it.
MCP The Facilities agent connects to the company calendar and room booking system through MCP servers. It checks Tuesday availability and finds three open rooms.
AG-UI As the agent works, it streams progress back in real time: "Checking calendar… Found 3 available rooms… Preparing options." The user sees each step as it happens.
A2UI The agent sends a surfaceUpdate with a room selection form—room dropdown, time confirmation, and a "Book" button. It renders as a native, styled form in the app.
The user picks a room and clicks Book. The action flows back through A2UI to the Facilities agent, which confirms the booking through MCP and returns a success artifact to the assistant via A2A. Done—five standards, one seamless experience.
Where the alternatives fit in: If the room booking API used UTCP instead of MCP, the agent would call it directly—no server needed, same result. If the UI was generated with Open-JSON-UI first, it would be translated to A2UI for rendering—simpler generation, same polished output.

Which One Do I Need?

Using AI coding assistants? → AGENTS.md
Connecting AI to tools & data? → MCP
Coordinating multiple agents? → A2A
Streaming agent responses live? → AG-UI
Generating rich UI from agents? → A2UI
Calling APIs directly, no middleware? → UTCP
The Seven Standards

Understand Each One

Tool Integration

MCP Widely Adopted

Model Context Protocol

The "USB-C for AI." One universal protocol that connects any AI model to any tool or data source. Instead of building custom integrations for every combination, everyone implements MCP once.

Tools Data Sources Universal
Configuration

AGENTS.md Easiest Start

Project Instructions for AI

A single Markdown file that tells AI coding assistants how your project works. What to do, what not to do, what commands to run. One file that works across 25+ AI tools.

Coding Zero Setup Vendor-Neutral
Agent Coordination

A2A Growing Fast

Agent-to-Agent Protocol

Lets AI agents from different vendors discover each other and work together on tasks. An agent built with one framework can hand off work to an agent built with another—no custom glue code.

Multi-Agent Collaboration Cross-Vendor
Real-Time Interaction

AG-UI Production Ready

Agent-User Interaction Protocol

Turns AI from a send-and-wait experience into a real-time conversation. Stream tokens as they're generated, watch tool calls happen live, pause and approve actions before they execute.

Streaming Events Human-in-the-Loop
UI Rendering

A2UI Public Preview

Agent-to-User Interface

Agents describe what they want to show, your app decides how to render it. Safe by design—agents can only use UI components you've approved. No code injection, no security risks.

Security Cross-Platform Declarative
UI Generation

Open-JSON-UI OpenAI Ecosystem

LLM-Optimized UI Format

A flat, simple JSON format that LLMs can generate reliably. Instead of complex nested UI specs that models get wrong, Open-JSON-UI keeps it simple—agents describe content, clients render it.

Token-Efficient Reliable LLM-Friendly
Direct Integration

UTCP Early Stage

Universal Tool Calling Protocol

Agents call your existing APIs directly—no wrapper servers, no middleware. If a human can call your API, an agent can too, using the same authentication and the same endpoints.

APIs Zero Overhead Low Latency