CopilotKit

🎉 28,000 GitHub Stars for CopilotKit! That means we are in the top .00017% of ALL repos Looking back: we started 2025 with 14k GitHub stars So it's great to open 2026 with DOUBLE that plus 11k stars for the AG-UI repo Everyone in the space likes to say GitHub stars are just a vanity metric... but even as we prioritize metrics more directly tied to real success (downloads, ARR, usage, etc), we’re seeing strong growth across all of them 💨 Our GitHub no longer serves as the main top of funnel to the product like it used to. These days, most traffic is funneled directly into the product & documentation. But I still love seeing it grow! --- I remember the days of getting excited about first few hundred stars as CopilotKit started resonating out there. And believe me, I don't take it for granted and STILL get excited every few hundred stars. Thank you to the tens of thousands of developers who have shown appreciation to our libraries! You are the ones who help us make CopilotKit and AG-UI better.

MCP ecosystem just got more practical. Until now, tools returned data and you had to rebuild the UX yourself (forms/tables/buttons) for every integration. MCP Apps (SEP-1865) make it possible for tool outputs to include interactive UI that the host app can render. But rendering UI is only step 1. The hard part is coordinating: - tool streaming + completion/failure - user interactions (clicks/submits) flowing back to the agent - agent state updates staying in sync with the UI I tried out CopilotKit’s new MCP Apps support (AG‑UI under the hood) and it basically acts as the runtime + coordination layer. With just a few lines of code, you can: - embed MCP-powered apps directly into your UI - stream interactive responses (not just text) - keep UI ↔ agent state synced out of the box You can get started with a single command: "npx copilotkit create -f mcp-apps" You can grab example MCP Apps servers from the official repo (modelcontextprotocol/ext-apps) and test it out. docs: https://lnkd.in/gYNTzGrE github repo (demo): https://lnkd.in/gTkJ65Tk blog: https://lnkd.in/gwbpFPHJ p.s. The blog goes into the MCP Apps spec and explains the full integration flow.

Introducing MCP Apps 🤝 AG-UI Bring MCP Apps to your own Agentic applications with a few lines of code We've been working for weeks now with the creators of MCP UI, Ido Salomon & Liad Yosef, to bring something very special for builders. And we're excited to finally share that you can now bring MCP Apps into your own agentic applications. 𝐖𝐡𝐚𝐭 𝐝𝐨𝐞𝐬 𝐭𝐡𝐢𝐬 𝐦𝐞𝐚𝐧? MCP Apps lets MCP servers return mini-apps to users via iFrames. It's been adopted by OpenAI and Anthropic for their upcoming app stores. With the AG-UI handshake, MCP Apps widgets become first-class citizens in your agent apps. Same runtime, same tools, same state. They're not just embedded; they're integrated. How it works: - 𝐓𝐨𝐨𝐥 + 𝐔𝐈: MCP servers expose tools that include a UI component - 𝐀𝐮𝐭𝐨-𝐫𝐞𝐧𝐝𝐞𝐫: When the agent calls the tool, the UI appears in chat automatically - 𝐓𝐰𝐨-𝐰𝐚𝐲 𝐜𝐨𝐦𝐦𝐮𝐧𝐢𝐜𝐚𝐭𝐢𝐨𝐧: The UI can call tools back on the server - 𝐀𝐠𝐞𝐧𝐭𝐬: Any AG-UI agent can now interface with an MCP app. Framework agnostic, supports HTTP and SSE 𝐒𝐞𝐞 𝐨𝐮𝐫 𝐌𝐂𝐏 𝐀𝐩𝐩𝐬 𝐝𝐞𝐦𝐨 → https://lnkd.in/ecKC2bVG CopilotKit is committed to providing developers with the go-to platform for building Agentic Applications & Generative UI. We're excited to keep collaborating with builders & ecosystems innovating in this space like MCP Apps. We will post a lot more about this in the year ahead! --- Read more: https://lnkd.in/ey_DVPCE Get started: https://lnkd.in/eQHcmc7z

2026 is the *real* year of agents Here's why: 1. There is a rise of insanely useful, general agents 2. There is infrastructure in place for anyone to build many more such agents 𝐔𝐬𝐞𝐟𝐮𝐥 𝐀𝐠𝐞𝐧𝐭𝐬: Claude Code, Manus, Claude Cowork: Millions of people are using these everyday for real productivity help, and they're real general agents. These agents have the right amount of generality & human-in-the-loop. I use these agents to automate parts of hiring, project management, mass reach outs, demo building and more. They really have crossed the threshold of 'cool demo' to actually useful. 𝐈𝐧𝐟𝐫𝐚𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞: The 'secret sauce' that made these agents useful is now known. On the backend side: planning mode, a file system, sub agents, system prompt. There is a great write-up on Deep Agents by Harrison Chase of LangChain: https://lnkd.in/ddqbNJYe On the frontend: Generative UI, great interactivity (allowing agents & human to collaborate), human-in-the-loop in the right places and agents <> application synchronization. Today we are releasing official support for LangGraph deep Agents inside of CopilotKit. We will do a lot more with this going forward. https://lnkd.in/dSa_rXVm --- So not only do we already have incredibly useful agents, used by millions daily, but the tools to build these agents for your use-case and in your environment - are readily available. 2025 wasn't "the year of agents", it was only the groundwork. 2026 is gonna be insane.

deepagents with beautiful frontends 👀 shoutout to our friends CopilotKit who’re always cooking at the frontier of generative UI love TUIs but more hype to see all the cool frontend homes ppl are gonna build for deepagents 🥹