Anthropic says Claude can now open and interact with third-party workplace tools inside the chat interface, starting with integrations that surface live, embedded UIs rather than text-only outputs.
In a January 26 product post, the company said users can draft and send Slack messages with a formatted preview, update timelines in Asana, and create diagrams in Figma “without switching tabs.”
The feature is accessed through an “interactive” apps directory, which Anthropic says is available on web and desktop for Claude’s Pro, Max, Team, and Enterprise plans.
The company also flagged a Salesforce integration as “coming soon,” positioning it as a way to bring additional enterprise context into Claude through Salesforce’s Agentforce 360.
Anthropic framed the product shift as a collaboration upgrade, arguing that Claude already “connects to your tools and takes actions on your behalf,” but that interactive apps make those actions visible in-context so users can “collaborate in real time.”
The initial list is enterprise-leaning: Box for file search and inline previews, Canva for deck creation and brand edits, Clay for prospecting and outreach drafting, plus analytics and project tools including Amplitude, Hex, and monday.com.
Built on an open standard shared with OpenAI
Anthropic is tying the rollout to the Model Context Protocol (MCP), an open standard it introduced in late 2024 for “secure, two-way connections” between AI tools and external data sources via MCP servers and MCP clients.
In the interactive-tools announcement, Anthropic said the new capability uses “MCP Apps,” an extension intended to let MCP servers deliver interactive interfaces inside “any supporting AI product—not just Claude.”
That “not just Claude” claim is doing strategic work. The MCP Apps Extension proposal (published in November 2025) explicitly builds on prior community efforts and references OpenAI’s Apps SDK work; it also notes the spec was authored by MCP core maintainers at both OpenAI and Anthropic alongside other contributors.
OpenAI, for its part, introduced “apps in ChatGPT” on October 6, 2025, saying the Apps SDK is “built on… the Model Context Protocol (MCP)” and that apps can include interactive interfaces inside chat.
The implication for developers and enterprise buyers is interoperability pressure: if interactive app delivery is standardized at the protocol level, the moat shifts from whether a model can connect to tools to how well the platform governs identity, permissions, and admin oversight across those connections.
Cowork is the autonomy amplifier, and governance is the constraint
Anthropic is also explicitly threading the new interactive tools into Cowork, its agentic “research preview” feature announced January 12.
Cowork gives Claude access to a user-selected local folder so it can read, edit, create and potentially delete files while it executes multi-step work with more autonomy than a standard chat. Anthropic said interactive apps are “coming soon” to Cowork, but are not integrated at launch.
This is where enterprise risk management becomes the gating factor, and Anthropic’s own documentation underlines it.
In its Cowork safety guidance, Anthropic urges users to avoid granting access to sensitive files (including financial documents, credentials, or personal records), to consider a dedicated working folder instead of broad permissions, and to monitor for prompt injection, especially when Cowork is pulling in untrusted web content or using unfamiliar MCP extensions.
For compliance-heavy environments, Anthropic’s Help Center also lists a hard limitation: Cowork activity “is not captured in Audit Logs, Compliance API, or Data Exports,” and it explicitly advises customers not to use Cowork for regulated workloads.
Taken together, those constraints clarify the near-term enterprise calculus: interactive tools can reduce workflow friction, but higher autonomy raises the bar for auditable controls and Anthropic is signaling that parts of its agent stack are still in preview-mode from a governance standpoint.
Why this fits Anthropic’s enterprise positioning now
Anthropic is launching the interactive directory as a paid-plan feature set and pairing it with an “open standard” narrative, two levers that matter in enterprise adoption cycles.
The timing also lands days after Anthropic published “Claude’s new constitution” (January 22nd), which argues that publishing a detailed values-and-behavior document improves transparency by helping outsiders distinguish intended from unintended behavior.
In enterprise settings where AI policy, procurement, and auditability increasingly move together, that transparency framing becomes part of the product story, not just the research story.