On January 26, 2026, Anthropic quietly changed what Claude is.
You're in a conversation, working through a project plan. You ask Claude to help you structure the timeline. Instead of getting a text outline, an Asana interface appears inside the chat. You drag tasks around, set dependencies, assign owners. The project exists in Asana now, visible to your team, built without leaving the conversation.
This is Claude apps. Not a plugin. Not a chatbot that talks to other software. Actual interactive tools running inside the conversation, responding to what you say and showing you interfaces you can use.
Anthropic is the second major AI platform to launch in-chat apps. OpenAI opened the ChatGPT App Directory a month earlier. But the way Claude apps work is different in ways that matter, especially if you're thinking about building for this ecosystem.
Here's what you need to know.
What Are Claude Apps?
Claude apps are interactive tools that render inside Claude conversations. They respond to natural language, display visual interfaces, and connect to external services. You can see them, interact with them, and take real actions without leaving the chat.
Ask Claude to help analyze your data, and an Amplitude chart appears that you can explore and adjust. Ask it to draft a Slack message, and you see a formatted preview you can edit before sending. Ask it to visualize an idea, and a Figma diagram takes shape while you watch.
The key word is interactive. These aren't static responses or text summaries of what an app could do. They're the apps themselves, running where the conversation happens.
What makes Claude apps architecturally interesting is how they're built. The foundation is MCP, the Model Context Protocol. Anthropic created MCP as an open standard for connecting AI to external tools and data. They open-sourced it, and now both Claude and ChatGPT build on it.
But Claude apps take MCP further. Anthropic introduced something called MCP Apps, a formal extension to the protocol that lets any MCP server deliver an interactive interface, not just data and actions. This isn't proprietary. Any AI product that adopts the MCP Apps standard can run these apps.
That's infrastructure thinking, not just product thinking. And it matters for anyone deciding where to invest their building efforts.
What's Available Today
Anthropic launched with a focused set of partners. The lineup tells you something about their priorities.
Amplitude lets you build analytics charts inside the conversation, then explore trends and adjust parameters to find insights. Asana turns chats into projects, tasks, and timelines your team can see and execute. Box lets you search files, preview documents inline, and ask questions about your content.
Canva creates presentation outlines you can customize with branding and design in real-time. Clay researches companies, finds contacts with email and phone info, pulls data like company size and funding, and drafts personalized outreach. Figma turns text and images into flowcharts, Gantt charts, and other visual diagrams in FigJam.
Hex lets you ask data questions and get answers with interactive charts, tables, and citations. monday.com manages work, runs projects, updates boards, assigns tasks, and visualizes progress. Slack searches and retrieves conversations for context, generates message drafts, and lets you format and review before posting.
Salesforce is coming soon, bringing enterprise context through Agentforce 360.
Notice the pattern. This is heavy on productivity and enterprise tools. Less consumer, less lifestyle. Anthropic is targeting work, and they're not being subtle about it.
Availability is limited to web and desktop for now, on Pro, Max, Team, and Enterprise plans. Apps are also coming to Claude Cowork, Anthropic's agentic desktop product. That's where things get more interesting for complex workflows, but we'll get there.
How Claude Apps Work
You don't need to understand the full technical spec, but the architecture explains why Claude apps matter.
MCP, the Model Context Protocol, defines how AI connects to external tools. An MCP server tells the AI what tools are available, what resources it can access, and what actions it can take. This is the plumbing that lets Claude talk to Asana or Figma or your company's internal systems.
MCP Apps extends this. Now an MCP server can also define interface components. When Claude determines an app is relevant to the conversation, it renders that interface inside the chat. You see buttons, forms, charts, previews. You interact with them directly.
The flow feels natural. You might say "help me plan this product launch" and Claude recognizes the context matches project management. It surfaces Asana. You build the plan in the interface that appears. The work happens in one place.
First-time connections prompt for authorization, so you control what data each app can access. After that, the app remembers your account and preferences.
The open standard angle matters here. Because MCP Apps is a protocol extension, not a proprietary SDK, apps built this way can theoretically run anywhere that adopts the standard. Today, Claude is the only place. But MCP is gaining traction. The investment you make in building an MCP app isn't locked to one platform.
How Claude Apps Compare to ChatGPT Apps
Both platforms now have in-chat apps. Both build on MCP. But the approaches differ.
ChatGPT's Apps SDK extends MCP with OpenAI-specific tooling. It's a more controlled environment. There's a formal App Directory, a submission and review process, and a clear path to getting your app in front of users. The ecosystem is a month older and more mature. Consumer apps like Spotify and Instacart sit alongside productivity tools.
Claude's approach leans harder on the open standard. MCP Apps is designed to be portable. There's no App Directory yet. Apps connect through Claude's existing connectors system. It's more open, but also less structured. The early partners are almost entirely enterprise and productivity focused.
For builders, the strategic question is which bet to make.
ChatGPT has more users and a head start on ecosystem development. If distribution is your priority, it's the obvious choice today.
Claude has a cleaner technical foundation and an explicit commitment to openness. If you believe MCP will become the industry standard, building for Claude means building portable assets.
The smart answer might be both. And that's where the tooling problem comes in.
Why Enterprises Should Pay Attention
Anthropic has always positioned Claude for enterprise. Security, compliance, thoughtful deployment. The Claude apps launch extends that positioning into workflow integration.
This matters for a few reasons.
First, if MCP becomes the standard, apps you build for Claude aren't trapped there. The protocol is open. Other AI products can adopt it. Your development investment has a longer shelf life.
Second, Anthropic's enterprise focus means B2B use cases won't be afterthoughts. The launch partners are tools enterprises actually use: Asana, Salesforce, Slack, monday.com. This isn't a consumer play with enterprise features bolted on later.
Third, the ecosystem is small right now. That's actually an advantage. Less noise. Easier to stand out. When the ChatGPT App Directory gets crowded, the early movers will have the featured placements and the user habits. Claude's ecosystem is at that early stage now.
The use cases are already compelling. Connect Hex or Amplitude and your team can ask data questions in natural language, getting interactive results instead of static reports. Use Asana or monday.com and project planning happens in conversation, with the artifacts living in tools your team already uses. The Clay integration turns Claude into a prospecting assistant that researches companies and drafts outreach without you copying data between tabs.
And then there's Cowork. Anthropic's agentic desktop product is getting apps too. That means Claude won't just surface interfaces in chat. It'll take actions on your computer, with app interfaces as the interaction layer. For complex enterprise workflows that span multiple systems, this is where the real potential lives.
How to Build Claude Apps
Here's the honest assessment of what building looks like today.
MCP Apps is a new extension to the protocol. Anthropic published the spec alongside the launch announcement. Documentation lives at modelcontextprotocol.io. It's real and it works, but it's early.
To build a Claude app, you need an MCP server that implements the Apps extension. That means defining what tools your app provides, what resources it can access, and what interface components it renders. You need a backend that handles authentication and data. And you need to connect it to Claude through their connectors system.
There's no formal "submit for review" process like ChatGPT has. The path is more open but also less structured. You're building on a protocol, not applying to a platform.
The friction points are real. Documentation is limited. The ecosystem is tiny, so there aren't many examples to learn from. There's no visual tooling. You're writing code to define interfaces, not designing them. And once your app is live, analytics and measurement are thin. You don't get much insight into how users interact with it inside Claude.
This is early-stage infrastructure. Powerful, but not polished.
The Case for Building Cross-Platform
Here's what's worth understanding: ChatGPT and Claude both build on MCP. The interface layer differs, but the foundation is shared. Building an app that works across both platforms is possible, and it's increasingly practical.
This matters because you don't have to pick a winner. You don't have to bet on Claude or ChatGPT and hope your choice ages well. Your investment in building an AI app can pay off across the ecosystem. As more AI products adopt MCP, your app's reach grows without you rebuilding it.
What makes this hard today is tooling. Each platform has quirks. Testing and deployment is manual. There's no unified way to design an interface once and ship it everywhere. If you're building from scratch with code, you're doing a lot of redundant work.
This is what we built Layo to solve.
Layo is a platform for designing, shipping, and measuring apps that live inside AI chat interfaces. You design in a visual studio. You deploy to Claude, ChatGPT, and other MCP-compatible platforms from one place. You get analytics that show how users interact across all of them.
If you're serious about building for this space, rebuilding for each platform doesn't make sense. Build once. Deploy everywhere. Iterate based on real data.
What's Next for Claude Apps
The launch is just the starting point. Here's what to expect.
Near-term, Anthropic will expand the partner list. More enterprise tools, probably deeper integrations with the launch partners. The Cowork integration will mature, bringing apps into agentic workflows where Claude takes action on your behalf.
At some point, there will be a discovery mechanism. An app directory or marketplace. Anthropic hasn't announced anything, but the pattern is obvious. When there are more apps than users can keep track of, you need a way to browse and search.
Monetization is the open question. OpenAI has talked about subscriptions, in-app purchases, and the Agentic Commerce Protocol. Anthropic hasn't announced anything on this front. But if they want developers to build seriously for Claude, they'll need to create revenue paths.
The bigger picture is the MCP bet. Anthropic is trying to make MCP the infrastructure layer for AI apps across the industry. If that works, Claude apps are portable and valuable. Builders win. If MCP stays niche, you're still building on solid technical foundations, but the cross-platform promise doesn't pay off.
Anthropic is playing a long game. The builders who understand that, and position accordingly, will have an edge.
The Opportunity Right Now
Claude apps just launched. The ecosystem is tiny. The tools are raw. There's no directory, limited documentation, and a small set of launch partners.
That's exactly when the opportunity is biggest.
The builders who start now will understand this space when everyone else is just arriving. They'll know what works in conversational interfaces. They'll have apps live and learning while competitors are still evaluating whether to build.
If you're thinking about AI apps for your enterprise, Claude deserves attention. The open architecture, the enterprise focus, the MCP foundation. There's a real case that this is the more interesting platform to bet on.
And if you want to build without choosing, if you want to ship to Claude and ChatGPT and whatever comes next, that's what Layo is for.