AG-UI: A Protocol to Connect AI Agents and Interfaces

A new protocol, AG-UI, organizes continuous conversation between an AI agent and the user. Supported by CopilotKit and designed to complement MCP and A2A, it relies on standardized events, a reference HTTP implementation, and an online accessible demo Dojo.
What AG-UI Promises and What It Does Not Deliver
AG-UI is presented as a means to facilitate the integration of agents into interactive applications and save considerable effort for developers. It does not aim to make agents smarter and does not replace existing agent frameworks. As agent-based AI matures, the interoperability promised by this type of protocol is touted as increasingly important. The stated goal of AG-UI is to meet the need for smooth interaction between an agent and the user.
A Unique Event Stream and 16 Standard Types
AG-UI is described as an event-driven protocol to connect agent back-ends to user-oriented applications. Designed to remain simple and adaptable, it relies on a unique event stream and 16 standard event types emitted during executions. Back-ends can also accept AG-UI compatible inputs as arguments. A flexible middleware layer ensures compatibility across various environments. The protocol comes with a reference HTTP implementation and a default connector, elements incorporated into the described architecture.
Complementarities and Limitations Compared to MCP and A2A
AG-UI has been designed to complement MCP and A2A. MCP, an open-source framework initially developed at Anthropic and later made available to the Agentic AI Foundation under the supervision of the Linux Foundation, facilitates integration with various sources, tools, and applications by reducing time and complexity, making it suitable for small-scale local projects. However, it is considered underperforming and insecure at scale, with a recommendation to use enterprise APIs for production use. A2A, an open standard launched by Google in April 2025 and integrated into the AAIF in August 2026, complements MCP without replacing it. This standard allows agents to identify each other's capabilities, negotiate exchange modalities (text, forms, media), cooperate securely on long-duration tasks, and operate without disclosing their internal state, memory, or tools. Technically, A2A is based on JSON-RPC 2.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Origins, Ecosystem, and Online Demonstrations
The protocol originated from an initial partnership between CopilotKit, LangChain, and CrewAI, before evolving to cover most major agent frameworks. In terms of adoption and examples, the AG-UI Dojo quickly grew, expanding from a set of four demos to over 30 in a week, although some older demonstrations, such as a Haiku generator, have disappeared. The Dojo is accessible online without prior installation and illustrates the joint functioning of the event stream, standard types, middleware, and HTTP implementation. The announced ambition is to reduce the need for custom integration code while supporting a wide variety of frameworks.
Why Agents Need an Interface Protocol
AI agents are described as having evolved from simple chats to systems used for serious work, with iteration cycles incorporating reflection, tools, analysis, and adjustments. They combine an LLM for reasoning, tools for action, short and long-term memory, and an orchestrator to guide the steps. Given this complexity, there is often a lack of a standard way to show the user what the agent is doing and to gather their actions without rebuilding infrastructure for each project. Several protocols already coexist for other needs, from commerce to payments and interface composition, while AG-UI is presented for continuous agent-user interaction. Demonstrations, including a previously documented Haiku generator, have been offered in an online space accessible without installation.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.