On January 26, 2026, Anthropic announced interactive connectors and MCP Apps for Claude. Workplace tools — Slack, Canva, Figma, Box, Clay and more — stop being APIs that return text and start opening inside the conversation as interfaces you can watch and operate. You can have Claude draft a Slack message and review it before it posts, or turn a rough paragraph into a FigJam flowchart, without ever leaving the chat window.
The real weight of the announcement sits in a second post published the same day on the Model Context Protocol blog: MCP Apps is now the first official MCP extension, which lets any MCP server deliver connectors with a rich user interface — and not only inside Claude. The chat window is turning from a question-answer surface into an application runtime, and for the first time that idea has a cross-vendor open standard underneath it.
What Interactive Connectors and MCP Apps Are
Under the old connector model, Claude read and wrote to external services through APIs and brought results back as text or files. You never saw an interface; you couldn’t reach in and adjust anything. Interactive connectors change that: external tools embed into the conversation as interactive UI components, and every step the model takes is visible and editable. Anthropic’s stated rationale is blunt — “analyzing data, designing content, and managing projects all work better with a dedicated visual interface.”
Each app provides a logged-in instance of the service, so Claude can search Slack conversations and format a message, adjust a deck’s branding in Canva in real time, or preview documents in Box. TechCrunch reports the feature is reaching paid plans first; users enable apps that carry an “interactive” label in the featured section of claude.ai/directory.
The Launch Lineup and What You Can Do
The nine launch apps are almost entirely workplace tools:
- Amplitude: build analytics charts, explore trends, adjust parameters live
- Asana: turn a chat into projects, tasks, and timelines
- Box: search files, preview documents, pull out the parts that matter
- Canva: generate presentation outlines and apply branding in real time
- Clay: research companies and contacts, pull funding data, draft outreach
- Figma: turn text and images into flowcharts and Gantt charts in FigJam
- Hex: data Q&A with interactive charts, tables, and citations
- monday.com: update boards, assign tasks, visualize progress
- Slack: search conversations, draft and format messages, review before posting
Salesforce is slated to join via Agentforce 360, bringing enterprise context into Claude. The detail worth watching is the relationship to Cowork, the general-purpose agent tool built on Claude Code that launched on January 12: at announcement time it could not use these apps, with Anthropic calling the integration “coming soon.” An agent that updates a Figma marketing graphic or reads from the company’s Box account is the next chapter of this playbook, not the current one.
The Protocol Layer: MCP’s First Official Extension
The technical route was proposed in November 2025, built on the open-source work of MCP-UI and the OpenAI Apps SDK, and shaped jointly by MCP maintainers, OpenAI, and the MCP-UI community. The mechanics, stripped down:
- A tool declares UI metadata in a _meta.ui.resourceUri field
- The UI resource is served by the server under the ui:// scheme, containing bundled HTML and JavaScript
- The host fetches it, renders it in a sandboxed iframe, and talks to the server over JSON-RPC via postMessage
The official SDK is the npm package @modelcontextprotocol/ext-apps, which provides an App class with methods like app.connect(), app.callServerTool(), and app.updateModelContext(). Apps can log events, open links in the user’s browser, send follow-up messages, and even update the model’s context in return. The example servers in the repo — threejs, map, pdf, system-monitor, sheet-music — are the fastest starting point for anyone building their own.
Beyond Claude: Build Once, Render Everywhere
The most consequential line in the MCP Apps announcement is the client list. Claude (web and desktop), Goose, and VS Code Insiders already support it; ChatGPT started rolling it out that same week. JetBrains, AWS (Kiro), and Google DeepMind (Antigravity) all said they will follow. OpenAI’s own Apps system shipped back in October 2025, but it is a walled garden; the MCP version is an open standard across clients — write the app once and, in principle, every supporting host can render it. David Soria Parra, MCP co-creator at Anthropic, said he is excited about what MCP Apps opens up, while OpenAI’s Nick Cooper said “we’re proud to support this new open standard.” When Microsoft, JetBrains, AWS, and Google all put their names on the same UI protocol, the interface-layer standards fight in the agent ecosystem is effectively over.
The Security Model and What It Means for Product Teams
Putting third-party UI inside a conversation raises the stakes, so the security design matters. MCP Apps runs UI exclusively in sandboxed iframes, requires pre-declared templates that hosts can review, keeps every JSON-RPC message auditable, and can require explicit user consent before a UI-initiated tool call executes. Anthropic’s own guidance in the Cowork docs is worth copying down verbatim: don’t grant permissions you don’t need, be cautious with sensitive material like financial documents, credentials, or personal records, and consider a dedicated working folder for the agent instead of broad filesystem access.
For product teams the signal is direct: the unit of competition in AI products is shifting from “model answers” to “app execution.” If your product already ships an MCP server, the next question is which operations deserve an upgrade to interactive UI — because the interface itself is now part of the protocol.
Sources
- Interactive connectors and MCP Apps — Anthropic
- Anthropic launches interactive Claude apps, including Slack and other workplace tools — TechCrunch
- MCP Apps: Bringing UI Capabilities to MCP Clients — Model Context Protocol Blog
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.