Anthropic updates MCP for custom UIs and long-running Claude tasks

Anthropic is bringing a major MCP update into Claude, and the useful part is not the protocol jargon. The change is that Claude connectors can become more like real product surfaces: they can show custom interfaces, run longer tasks and connect through enterprise identity systems.

MCP Apps let a connected server render interactive UI directly in the Claude conversation. That means a connector does not have to return only text or links; it can show a small interface for reviewing, editing or acting on results without sending the user to another tab.

MCP Tasks give developers a cleaner way to handle work that does not finish in one quick request. Instead of forcing everything into a short tool call, a connector can represent longer-running jobs more naturally.

Anthropic is also emphasizing enterprise-managed auth and observability for published connectors. For companies, the value is simpler rollout: admins can approve access once, users inherit it through existing identity groups, and developers can see how their connectors perform across Claude surfaces.

The practical takeaway: MCP is moving from “AI tool calling” toward a standard for richer apps inside AI assistants. If this works, Claude can become a place where people do work with connected tools, not only ask an assistant to summarize them.

Sources:

• https://claude.com/blog/bringing-mcp-2026-07-28-to-claude