Anthropic Brings New MCP Capabilities to Claude: What Changed

What are the new MCP capabilities in Claude

Anthropic's new MCP capabilities in Claude come from MCP 2026-07-28, the fifth major spec release of the Model Context Protocol, published on July 28, 2026. The update adds interactive tools inside Claude's chat window, support for long-running background work, and enterprise-grade authentication, on top of a rebuilt protocol core.

MCP itself is the open standard Anthropic introduced to let AI models like Claude connect to outside tools and data sources, such as a company's file storage or project tracker, through a common interface instead of a custom integration for every service. The protocol has grown fast: SDK downloads reached 400 million a month, roughly four times where they stood a year earlier.

From stateful to stateless

The biggest structural change is under the hood. Earlier versions of MCP required a persistent, bidirectional connection between Claude and each server it talked to, which made servers harder to scale and awkward to run on serverless platforms. MCP 2026-07-28 replaces that with a stateless request/response core, so a server can respond to a request and shut down instead of holding a connection open.

Interactive apps inside Claude

The most visible change for everyday users is MCP Apps, an extension that lets an MCP server render its own interface directly inside Claude's chat, instead of just returning text. Early partners include Asana, Box, Canva, Figma, Hex, monday.com, and Slack, so a user can update a project timeline or edit a design without leaving the conversation.

This fits a broader shift already underway in how people interact with AI tools, where conversational interfaces are replacing rigid chatbot boxes in favor of something closer to a real workspace. MCP Tasks, a second extension shipping alongside MCP Apps, handles work that takes longer than a single request and response, letting a server report progress on a long-running job instead of timing out.

Enterprise authentication

For companies rolling this out to employees, the update's authorization changes matter as much as the interactive apps. MCP servers can now plug into identity providers such as Entra or Okta using standard OAuth 2.0 and OIDC flows, which Anthropic describes as removing the need for custom workarounds that many enterprise deployments previously required.

That change lines up with the same zero-trust thinking enterprises are already applying elsewhere, where every connection needs its own verified identity rather than relying on network location for trust. Anthropic also added developer observability dashboards and MCP tunnels, currently in research preview, for reaching servers on private networks.

What this means for developers

For teams building on MCP, the stateless core is the practical win: servers can now deploy on the same serverless and edge infrastructure already used for the rest of a typical stack, instead of requiring dedicated always-on hosts. That lowers the operational cost of shipping a new connector and should accelerate the kind of agentic workflows that depend on Claude reaching many small tools reliably.

Claude's connectors directory has grown to more than 950 MCP servers alongside this release, with partners including Figma, Intuit, Netlify, PostHog, Xero, and Zoom already building on the new spec. For developers deciding whether to invest in an MCP server versus a one-off API integration, the stateless model and formal extensions framework make MCP a materially safer long-term bet than it was under the previous spec.