AG-UI (Agent-User Interaction Protocol)
AG-UI (Agent-User Interaction Protocol) official site ↗Last checked against the official docs:
What it does
An open, event-based protocol for connecting AI agents to the apps people use. Instead of custom wiring for every framework, the agent streams typed events (text as it is written, tool calls, shared state updates, pauses for a human) over one stream, and any AG-UI client can show them. It sits next to MCP (which gives agents tools) and A2A (agent to agent): AG-UI is the agent-to-user part. Version 1.0 has a stable specification, with TypeScript, Python and .NET SDKs, and integrations for frameworks such as LangChain, CrewAI, Mastra, Pydantic AI, Google ADK and the Claude Agent SDK. CopilotKit is one of the frontend clients. MIT licensed.
Who it's for
Developers building their own AI agent who want to put it in a web app, mobile app or chat tool without writing a custom streaming layer for each one.
How to set it up
Start a new app: the CLI scaffolds a frontend (CopilotKit on Next.js, or a terminal client) wired to the agent framework you choose. The generated templates call OpenAI, so set OPENAI_API_KEY before running npm run dev:
bashnpx create-ag-ui-app@latestOr speak the protocol directly. The agent side is any HTTP endpoint that accepts a run request and streams events back. Python SDK (Python 3.9 or later):
bashpip install ag-ui-protocolTypeScript client (Node.js 20 or later):
bashnpm install @ag-ui/clientThe README has a short working server and client example, and the Dojo (dojo.ag-ui.com) has live demos for each supported framework.
Check the official AG-UI (Agent-User Interaction Protocol) docs for the latest steps.
Example prompts
Add an AG-UI endpoint to my FastAPI agent so it streams text and tool calls to the frontend.Connect this Next.js app to my LangGraph agent through AG-UI and show tool calls as they happen.Explain which AG-UI events I need to support a human approval step before the agent sends an email.Made AG-UI (Agent-User Interaction Protocol)? Add the badge
If you like, show that AG-UI (Agent-User Interaction Protocol) is listed and hand-checked here. It's optional: listings are free and we never ask for a link back.
Get the Markdown or HTML
Markdown (for a GitHub README)
[](https://www.pluginstackai.com/plugins/ag-ui/)HTML (for a website)
<a href="https://www.pluginstackai.com/plugins/ag-ui/"><img src="https://www.pluginstackai.com/badge/hand-checked.svg" alt="Hand-checked on PluginStack" width="226" height="28"></a>Free: MCP Starter Checklist
After you set this tool up, use the one-page checklist so your next MCP server doesn't fail on first connect.
More Agent Frameworks
- Vercel
AI SDK
Vercel's provider-agnostic TypeScript toolkit for building AI apps and agents. It works with Next.js, React, Svelte, Vue, Angular and Node.js, and its UI hooks cover React, Svelte and Vue. One API covers OpenAI, Anthropic, Google and many more providers, either through Vercel AI Gateway (the default) or each provider's own package. Apache-2.0.
- CrewAI
CrewAI
A Python framework for multi-agent automation. Crews are teams of role-based agents that work together on a task; Flows are event-driven workflows with state and branching that can call crews. The CLI scaffolds projects, which are JSON-first by default. MIT licensed.
- LangChain
LangGraph
A low-level framework for building long-running, stateful agents as graphs. It provides durable execution (agents resume after failures), human-in-the-loop interrupts and short- and long-term memory. It works on its own or with LangChain and LangSmith. A JS version (LangGraph.js) is also available. MIT licensed.
- Mastra
Mastra
A TypeScript framework for AI apps and agents. It includes model routing across 40+ providers, agents with tools, graph-based workflows with suspend/resume, memory and RAG, MCP server authoring, evals and observability, plus a local Studio UI. Apache-2.0, except code in ee/ folders, which uses an enterprise license.
- OpenAI
OpenAI Agents SDK
OpenAI's lightweight, open-source framework for building agents and multi-agent workflows, in Python (openai-agents) and JavaScript/TypeScript (@openai/agents). It covers agents with tools (functions, MCP servers, hosted tools), handoffs, guardrails, human-in-the-loop, sessions, tracing, sandbox agents and realtime/voice agents. Per the README it is provider-agnostic: it supports the OpenAI Responses and Chat Completions APIs and 100+ other LLMs. MIT licensed.
- Pydantic
Pydantic AI
A typed Python agent framework from the Pydantic team. You define agents with typed outputs and tools, and switching model providers is a one-string change. Companion packages add a harness for long-running agents (files, shell, sub-agents) and Pydantic Evals for testing. MIT licensed.