Anthropic is turning Claude into something that looks a lot less like a chatbot and a lot more like a full-blown workspace. Instead of just asking the model to “send a Slack message” or “make a deck in Canva” and then jumping over to those apps yourself, you can now actually open and use those tools right inside the Claude interface.

On the surface, the update sounds simple: Claude now supports “interactive apps” for a growing list of popular work tools. At launch, that list covers Amplitude, Asana, Box, Canva, Clay, Figma, Hex, monday.com, and Slack, with Salesforce integrations (like Agentforce 360) waiting in the wings. If you’re on a paid Claude plan on web or desktop, you can head to the Claude app directory, flip on any connector marked “interactive,” and those tools will start showing up directly in your chats as live, clickable UIs rather than just text responses.

In practice, this means your conversations with Claude don’t have to stop at “here’s a suggestion” — they can flow straight into execution. Ask Claude to turn a messy project brief into a real plan, and it can spin that into a structured Asana board with timelines and tasks that you can review and tweak without leaving the chat. Tell it to draft a Slack announcement for your product launch, and you’ll see a proper Slack composer embedded in the conversation, with formatting, channels, and previews you can adjust before hitting send. Want to turn a brainstorm into something visual? You can prompt Claude to generate a FigJam diagram in Figma — flowcharts, Gantt charts, system diagrams — and then refine it directly from the same thread.

The rest of the lineup follows the same pattern: Hex and Amplitude can surface live charts and tables you can probe and adjust, Box can show inline previews of documents you’re asking Claude about, Canva can go from “outline this deck” to an actual editable presentation, Clay can handle company research and outreach drafts, and monday.com becomes a living project board that sits in the middle of the chat. The connective tissue here is that Claude isn’t just summarizing what these tools might do; it’s effectively pulling their mini-interfaces into the conversation so you can steer both the AI and the app at the same time.

Under the hood, this all rides on Anthropic’s Model Context Protocol (MCP), and more specifically, a new extension called MCP Apps. MCP started life as an open standard for letting AI agents talk to tools — fetch data from your CRM, read from your analytics stack, hit internal APIs — and Anthropic has now pushed it a step further by letting those tools render real UI components right in the chat window. Think dashboards, forms, multi-step workflows, data visualizations: instead of getting a wall of JSON or text back, you see an actual mini-app that you can click and manipulate, with your actions feeding back into the model’s context.

There’s also a strategic angle: by open-sourcing MCP and positioning MCP Apps as an official extension, Anthropic is betting that developers won’t just build “Claude integrations,” they’ll build MCP apps that can plug into any MCP-capable client — whether that’s Claude, a future ChatGPT client, or something else entirely. MCP Apps exposes a simple API that lets an app render UI, receive tool results, send events back, update the model’s context, and generally behave like a first-class panel inside an AI chat, all over standard web messaging. That lowers the friction for tool builders and quietly nudges the ecosystem toward a shared way of embedding interactive experiences into AI assistants.

For users, the impact is more tangible than the protocol talk might suggest. If you already live in Slack, Figma, Asana, and Canva all day, this is an attempt to make Claude the central glass pane where those workflows start, get orchestrated, and often get finished. You could imagine: planning a marketing campaign with Claude, having it pull historical performance charts from Amplitude, drafting copy variations, pushing tasks into monday.com, generating creative directions in Canva, and then dropping final announcements into Slack — all without context-switching through a dozen tabs.

It also changes the feel of “AI agents” in a subtle way. Instead of invisible background automations that fire off in tools you never see, the work is happening in front of you, in components you can poke, correct, or override. That’s partly a usability win — you see what’s going on — and partly a safety story, since humans stay in the loop as the agent touches critical systems like project management, data analytics, or messaging.

Of course, this also puts Anthropic in more direct competition with other “AI as workspace” plays. OpenAI is building its own app ecosystem and agentic flows, Microsoft is infusing Copilot across the Microsoft 365 universe, and Google is wiring its Gemini models into Workspace. Anthropic’s twist is to flip the model: instead of sprinkling assistants into every app, it’s trying to pull apps into one assistant, with MCP as the common glue.

Right now, interactive Claude apps are available on web and desktop for Pro, Max, Team, and Enterprise plans, with Cowork — Anthropic’s agent-style automation layer — set to tap into the same system soon. If Anthropic can get developers to embrace MCP Apps and enterprises to trust embedded third-party UIs inside their AI clients, Claude’s chat window could quietly become the place where a lot of company work actually happens.