Anthropic’s model context protocol (MCP), the ‘plug-and-play bridge for LLMs and AI agents’ to connect with external tools, has received a major update one year after its launch.

The developer of Claude is adding functionality for more complex workflows and offering support for task-based workflows, while simplifying authorization flows in the tool.

After its launch in November 2024, MCP was rapidly adopted by the AI industry as a tool to connect AI to database systems, in a bid to facilitate AI agents' access to tools and services and improve the way they communicate.

Industry giants like Microsoft have already added native MCP support to Windows 11, as the open ecosystem on GitHub has also been filled with reference servers and clients, indicating a fast-maturing standard.

One of the biggest changes as part of the update is task-based workflows, now offered experimentally. This allows a user to kick off long-running operations on an MCP server and later check their status or retrieve results for processes lasting minutes or even hours.

Authorization has also been simplified. Instead of requiring pre-registered users via a central Authorization Server, MCP now supports URL-based client registration. A user can provide a URL pointing to a metadata JSON document describing its credentials, which streamlines the onboarding of new users.

Another feature – sampling with tools – lets MCP servers themselves run agentic loops using a user’s tokens while preserving user control for reduced complexity.

The update further bolsters security and enterprise readiness. The anniversary refresh adds stricter user security requirements for local server installations, defines default authorization scopes, and enables enterprise identity-provider (IdP) policy controls and OAuth client-credentials flow for machine-to-machine authentication.

The prominence of Model Context Protocol

While AI assistants are already serving an important role in connecting to services through tools and API calls, they often require custom integrations for proprietary connectors such as APIs. The need to build custom code for each new service or data source creates hurdles for developers, as minor tweaks to one API could break the entire chain.

MCP addresses this issue by providing a universal interface for translating different tools and services in a way that the AI model can understand. Regardless of vendors, platforms, or unique APIs, the model can retain context with MCP.

Amid MCP’s quick proliferation, industry experts are somewhat divided on whether MCP will become the default agent interface in future networks. While some envision it becoming the universal translator for network teams between AI tools and infrastructure, others are not convinced it will see mainstream adoption, even after full development, with concerns for both increased technical overhead and security risks.

Commenting on the anniversary, Satyajith Mundakkal, global CTO of Hexaware Technologies, said: “AI is only as good as the data it can reach safely. That’s where MCP is proving essential – a standardized, governed way for assistants and agents to securely discover, request, and use enterprise data and tools. Done right, it delivers precision and trust at the same time – richer context for models, with policy controls, auditability, and least-privilege access baked in.”

However, Mundakkal warned that if done poorly, MCP becomes integration sprawl and a bigger attack surface.

“The lesson from year one is clear: pair MCP with strong identity, RBAC (role-based access control), and observability from day one,” he added.