Anthropic expands Claude connectors to open full enterprise apps inside chat

Anthropic has significantly expanded the capabilities of its Claude AI assistant to interact directly with enterprise software interfaces. Announced on November 10, 2026, at The Briefing: Work at the Frontier in San Francisco, the update allows users to open full applications such as Adobe, Atlassian, and HubSpot within the Claude conversation window. This move shifts the AI from a passive information retriever to an active agent capable of executing tasks across multiple platforms while adhering to strict permission controls.

What happened

The core of this update is the introduction of enhanced interactive connectors built on MCP Apps. Previously, users could only see small previews of connected applications. Now, Claude can render and interact with the full application interface inside the chat. For instance, a user can ask Claude to edit a Confluence page or update an Adobe design using specific brand assets, and the AI will perform these actions within the embedded interface. Crucially, users must still approve any changes before Claude executes them, maintaining a human-in-the-loop safeguard.

These connectors are not exclusive to Claude. Because they are built on the Model Context Protocol (MCP) Apps standard, other AI clients that support MCP can also utilize these interactive interfaces. This openness allows developers to bring existing application interfaces into AI assistants without rebuilding them from scratch. Anthropic stated that the Claude Marketplace now hosts more than 9,000 connectors and plugins, facilitating deeper integration with enterprise infrastructure. For applications lacking a dedicated connector, Claude can fall back on its built-in browser to interact with web-based tools.

Alongside the connector updates, Anthropic introduced several new features aimed at enterprise workflows. Claude Data, formerly known as Claude Dashboards, is now generally available with a new "certified queries" feature. This allows data teams to approve specific queries used for reporting, ensuring consistency across different departments. Additionally, beta features like Claude Meetings, Claude Whiteboards, and Claude Sheets are being rolled out to help teams collaborate on project plans and spreadsheets directly within the AI environment. However, some advanced features like Claude Meetings will not be available to organizations using zero data retention or customer-managed encryption keys.

Key details

• Full app integration: Users can open and work in full versions of Adobe, Atlassian, and HubSpot apps directly inside Claude, rather than just seeing previews.
• MCP Apps foundation: The connectors use MCP Apps, allowing any AI client supporting the protocol to use these interactive interfaces, not just Claude.
• Permission preservation: Connectors respect existing user permissions; if a user cannot access a file in Google Drive, Claude cannot retrieve it for them.
• Admin controls: Administrators can restrict connector capabilities, such as allowing read-only access to Slack without permitting message sending, and use "Claude auto-connect" for easier setup.
• Certified queries: Claude Data now supports approved queries to ensure consistent reporting numbers across teams, though versioning details remain unclear.
• Enterprise adoption: Mercado Libre reports deploying Claude to over 26,000 employees, with 78% weekly usage, although these figures are self-reported.

Background

To understand the significance of this update, it helps to know what MCP (Model Context Protocol) is. MCP is an open standard that allows AI models to connect securely to data sources and tools. Think of it as a universal USB port for AI, letting different software systems plug into an AI assistant without custom coding for each one. MCP Apps extend this by allowing the AI to not just call functions but to display and interact with the actual user interface of those tools. This reduces the friction of switching between tabs and applications, aiming to create a unified workspace where the AI acts as the central operating system for enterprise tasks.

Why it matters

For IT leads and engineering managers, this shift represents a major change in how software security and access control must be managed. The promise that "existing permissions still apply" is critical. If an employee does not have access to sensitive HR records in your primary database, Claude should not be able to bypass that restriction via a connector. However, verifying that these permission layers hold up across thousands of connectors requires rigorous testing. IT teams will need to audit which connectors are enabled and ensure that the "human-in-the-loop" approval process is enforced for high-risk actions like deleting data or sending external communications.

Furthermore, the move toward "certified queries" addresses a common pain point in business intelligence: inconsistent data. When different teams ask an AI for revenue figures, they often get different answers because the underlying queries differ. By allowing data teams to certify specific queries, organizations can trust that the AI is pulling from a single source of truth. This is particularly relevant for companies running their own software stacks, where data integrity is paramount. It suggests a future where AI assistants are not just creative tools but reliable operational partners that adhere to strict governance standards.

What you can do

• Audit current connectors: Review the 9,000+ connectors in the Claude Marketplace to identify which ones your team actually needs and disable unused ones to reduce attack surface.
• Test permission boundaries: Conduct internal tests to verify that Claude respects role-based access controls. Try accessing restricted files through the AI to ensure the security layer holds.
• Implement certified queries: If you use Claude Data, work with your data team to define and approve a set of standard queries for key metrics like sales or uptime.
• Configure admin restrictions: Use the new admin controls to limit connector capabilities. For example, allow read-only access to communication tools like Slack until you are confident in the AI's judgment.
• Evaluate MCP compatibility: If you develop internal tools, consider building them with MCP Apps support so they can be easily integrated into Claude or other compatible AI assistants in the future.
• Monitor usage with Smart Reports: Enable Smart Reports to gain visibility into how teams are using connectors, tracking completed work and spending to optimize license allocation.