Onpode
Cover art for Claude agents just became AG-UI compatible, expanding deployment options for developers

Claude agents just became AG-UI compatible, expanding deployment options for developers

August 13, 2026 · 9 min

Spuds Oxley

Claude Managed Agents gained full AG-UI protocol support on August 12, 2026 — announced by CopilotKit's CEO, not Anthropic. The integration, built with an Anthropic employee, eliminates the custom transport and state-sync code developers previously wrote from scratch, enabling React, Next.js, Angular, Slack, and Teams deployments out of the box.

On or around August 12, 2026, Anthropic shipped a full AG-UI integration for Claude Managed Agents (CMA), enabling the hosted agent runtime to connect to AG-UI-compatible frontends and exchange streaming events, tool calls, and state updates without custom glue code.

0:009:27
Get the next episode on Claude

Follow it free — new episodes land in your feed.

Or make your own — any topic, in minutes

More Onpode episodes on Claude

About this episode

On August 12, 2026, CopilotKit's CEO posted a demo video stating that Claude Managed Agents now have full AG-UI support. 132 likes, 12 reposts. No official Anthropic press release followed. No release notes dated that week. The primary evidence is a third-party blog post and a single social post, despite an Anthropic employee being credited on the integration itself. This episode works through what that actually means — technically and structurally. AG-UI is the Agent–User Interaction Protocol, an open, event-driven spec designed to do one thing: standardize how a running agent streams real-time events to a frontend. Not tool connectivity (that's MCP), not agent-to-agent routing (that's A2A) — the layer in between those two and the screen a person is actually looking at. That gap was real, and developers were closing it manually every time they shipped. The episode is honest about what's confirmed and what isn't. Framework parity across React, Angular, and the Channels SDK is unverified. The sentiment in the developer community was uniformly positive the week of the announcement — but a small room agreeing with itself and genuine consensus look identical from the outside, right up until they don't. What makes this worth your time isn't the protocol spec. It's the question underneath: can a third-party platform's credibility carry something into standard status without the official word ever coming? It has happened before. Whether it's happening here is a specific, observable thing — and the episode tells you exactly what to watch.

Frequently asked

What is AG-UI and how does it relate to Claude agents?

AG-UI (Agent–User Interaction Protocol) is an open, event-driven protocol that standardizes how a running AI agent streams real-time events to a frontend. Claude Managed Agents, Anthropic's hosted runtime for long-horizon tasks, became AG-UI compatible in August 2026, filling the missing layer between agent runtime and user-facing interfaces.

What did CopilotKit's Claude AG-UI integration actually change for developers?

Before CopilotKit's Claude Managed Agents AG-UI integration, developers hand-built every transport layer, state synchronization mechanism, and UI primitive connecting CMA to a frontend. The August 2026 integration replaces all of that custom glue code with an open-source, drop-in solution supporting React, React Native, Next.js, Angular, Slack, and Microsoft Teams.

Did Anthropic officially announce Claude Managed Agents AG-UI support?

Anthropic did not publish an official press release or release notes confirming Claude Managed Agents AG-UI compatibility as of the announcement date. The news arrived via a CopilotKit CEO social post on August 12, 2026, and a CopilotKit blog post. An Anthropic employee, @cjav_dev, is credited on the integration but no official Anthropic documentation has confirmed it.

What is the difference between AG-UI, MCP, and A2A protocols for AI agents?

MCP handles tool connectivity for AI agents, A2A handles agent-to-agent routing, and AG-UI handles the distinct layer between those two and the actual user-facing screen — specifically, streaming real-time events from a running agent runtime to a frontend. AG-UI was designed to fill the gap neither MCP nor A2A addressed.

Which frontend frameworks and platforms does Claude Managed Agents AG-UI support cover?

According to CopilotKit's blog post announcing the integration, Claude Managed Agents AG-UI compatibility covers React, React Native, Next.js, and Angular on the web side. Via the Channels SDK, it also extends to Slack and Microsoft Teams. Whether Angular and Teams reach full feature parity with React — including generative UI and tool call state — has not been independently confirmed.

Grounded in 6 sources
AG-UI: the Agent-User Interaction Protocol. Bring ... · github.com
GitHub - nanocoai/nanoclaw: A lightweight alternative to OpenClaw that runs in containers for security. Connects to WhatsApp, Telegram, Slack, Discord, Gmail and other messaging apps,, has memory, sch · github.com
Yeachan-Heo/oh-my-claudecode: Teams-first Multi-agent orchestration for Claude Code · github.com
ai-boost/awesome-harness-engineering · github.com
CopilotKit/channels-sdk: The open-source ... · github.com
Claude Managed Agents are now AG-UI compatible | Blog · copilotkit.ai
Read transcript

Spuds Oxley: The announcement that changes how Claude agents reach users — it didn't come from Anthropic.

Spuds Oxley: A hundred and thirty-two likes on a social post. Twelve reposts. That's the footprint. CopilotKit's CEO, @ataiiam, August 12, 2026 — posts a demo video and just states, plainly, that Claude Managed Agents now have full AG-UI support.

Spuds Oxley: Claude Managed Agents — CMA — that's the hosted agent harness Anthropic built for long-horizon tasks. The server-side loop that holds a session open, runs tools in a sandbox, streams results back. It's the runtime.

Spuds Oxley: AG-UI is what it now speaks.

Spuds Oxley: The Agent–User Interaction Protocol — event-driven, open, lightweight — its job is to standardize exactly one thing: how a running agent streams real-time events to a frontend. Not tool connectivity, that's MCP. Not agent-to-agent routing, that's A2A. The piece between those two and the actual screen a person is looking at. That was the missing layer. AG-UI is it.

Spuds Oxley: And developers connecting CMA to an AG-UI frontend used to build every part of that bridge themselves. Transport layer, state synchronization, UI primitives — written from scratch, every time.

Spuds Oxley: CopilotKit's blog post says that custom glue code is now gone. The integration — built with Anthropic's @cjav_dev — replaces all of it, out of the box, open source.

Spuds Oxley: And honestly, the fact that this came from a third-party platform and a single social post rather than any official Anthropic release — that's not a footnote. That's the story.

Spuds Oxley: Think about what a developer was actually doing before this.

Spuds Oxley: You've got Claude Managed Agents running the loop server-side — session is open, it's holding the transcript, executing tools inside a sandbox, streaming events back to the client. The agent itself is the model, the system prompt, the tools, the MCP servers, the skills — all of that wired together, managed by Anthropic's infrastructure.

Spuds Oxley: And then — on the other end — is a React app. Or a Next.js surface. Or a Slack channel.

Spuds Oxley: That gap between CMA and whatever the person is actually looking at — that was yours to close.

Spuds Oxley: Transport layer. State synchronization. UI primitives for streaming tool calls, generative UI updates, session lifecycle events — you built that. From scratch. Every time you shipped.

Spuds Oxley: That work is now zeroed out.

Spuds Oxley: The CopilotKit blog post says it plainly — drop Claude's hosted runtime into an existing AG-UI app, no custom bridges. Generative UI, streaming, tool calls, stateful sessions — out of the box. React, React Native, Next.js, Angular on the web side. And via the Channels SDK, Slack and Microsoft Teams. All of it open source.

Spuds Oxley: That's a real change in who can build this, and how long it takes them.

Spuds Oxley: But the announcement itself is doing something.

Spuds Oxley: Anthropic built this. @cjav_dev — an Anthropic employee — is credited on the integration. And yet the news arrived via @ataiiam, CopilotKit's CEO, on August 12, in a demo video with 132 likes and 12 reposts.

Spuds Oxley: One research source went looking for an official Anthropic announcement — a press release, release notes, documentation dated August 13, 2026. Found nothing. The primary evidence is the CopilotKit blog post and that single social post.

Spuds Oxley: So 'official' is doing complicated work here.

Spuds Oxley: And I don't think that's a press strategy failure. I think it's just — this is how protocol adoption actually propagates. Not through announcements. Through the people who have skin in the game, who've already integrated, who need the ecosystem to follow them.

Spuds Oxley: CopilotKit has platform reasons to make this work. They surface the integration because they ARE the integration.

Spuds Oxley: The AG-UI community has been saying for months that neither MCP nor A2A was designed for this — the real-time streaming layer between agent runtime and the actual frontend. MCP handles tool connectivity. A2A handles agent-to-agent routing. AG-UI is the piece in between, the event stream from the running agent to the screen. That was the gap. And the protocol that fills a gap doesn't need a keynote — it needs someone to ship against it.

Spuds Oxley: That tension — between unverified and consequential — is exactly where this story lives.

Spuds Oxley: Here's what I actually want to watch — and I mean watch, the way you watch a building go up, checking whether the frame holds.

Spuds Oxley: The social signal search — S05, Grok, the week of August 12, 2026 — came back with zero pushback. Zero fragmentation concerns. No competing protocol debates. Sentiment uniformly positive, across every expert account that met quality filters.

Spuds Oxley: Which could mean consensus.

Spuds Oxley: Or it could mean this was a very small room, and the room agreed with itself because not enough people had found the door yet. Those two things look identical from the outside, right up until they don't.

Spuds Oxley: And there are specific questions under that — concrete ones, not philosophical ones. Angular versus React. The blog post names all four frameworks: React, React Native, Next.js, Angular. But naming them and shipping them at the same level of maturity are not the same thing. React almost certainly got there first. Angular — I don't know. Nobody has said.

Spuds Oxley: The Channels SDK is another open edge. Slack and Teams are listed. How complete that is in practice — whether a developer routing a Claude Managed Agents session through Teams actually gets generative UI and tool call state the same way a React app does — that has not been confirmed anywhere I can find.

Spuds Oxley: These matter. Not abstractly.

Spuds Oxley: Because if AG-UI is going to be the standard — the ACTUAL standard, not just CopilotKit's preferred layer — developers who already know how to build agents without it have to decide to route through it anyway. Someone who can wire their own transport, sync their own state, write their own UI primitives — they have to look at this and decide the protocol is worth the dependency. That's the real adoption test.

Spuds Oxley: And the thing I keep returning to is this: the integration's legitimacy, right now, rests entirely on CopilotKit's blog post and @cjav_dev's social credit. If Anthropic never publishes release notes — never officially confirms that Claude Managed Agents and AG-UI are compatible — then this lives or dies by ecosystem adoption alone.

Spuds Oxley: That is a specific, observable thing to track.

Spuds Oxley: Watch Anthropic's documentation. Not the blog, not a social post — the actual release notes, the official CMA pages. If they update to name AG-UI compatibility, the story changes shape. If they don't — if this stays a third-party announcement backed by one Anthropic employee's handle — then what we're really watching is whether CopilotKit's credibility can carry a protocol into standard status on its own.

Spuds Oxley: That has happened before. It is not impossible. But it is a different kind of standard — one that holds as long as the people who need it keep building against it, and no further.

Spuds Oxley: Standards don't win because they're announced. They win because someone else builds on them first — because the people with platform reasons, with integration reasons, with skin-in-the-game reasons, they ship against it before the official word ever comes. And then the official word either follows, or it doesn't, and by then it may not matter.

Spuds Oxley: CopilotKit announced this. Not Anthropic.