The Open Source model context protocol has just been updated -here it is why it is a big deal

The model context protocol (MCP) – an increasing open standard, which is intended to help AI agents interact seamlessly with tools, data and interfaces – only affects a significant milestone. Today, developers behind the initiative have completed an updated version of the MCP specification, in which important upgrades were introduced to make AI agents safer, capable and interoperer.

In a very important step, Openai, the industry leader in generative AI, followed the MCP announcement today by saying that he also adds support for MCP for his products. CEO Sam Altman said the support was available today In Openais agents SDK and this support for the Desktop -app from Chatgpt and the API of the answers will take place shortly.

Microsoft Announced support for MCP in addition to this publication, including the start of a new one Dramatic MCP server As a result, AI agents such as Claude can search on the Internet and interact with websites with the Chrome barrier -freiberbaum.

“This new version is an important leap for agent tool communication,” said Alex Albert, an important contribution to the MCP project, said In a post On Twitter. “And Microsoft builds real infrastructure, shows how quickly this ecosystem will develop.”

What is new in the updated MCP Version?

The update of March 26 brings several important changes at the protocol level:

• Oauth 2.1-based authorization framework: Add a robust standard for securing communication to the agent server, especially in HTTP-based transports.
• Streamer HTTP transport: Replaces the older HTTP+SSE setup and enables the bidirectional data flow in real time with better compatibility.
• JSON-RPC-Batching: Allows clients to send several inquiries at once, to improve efficiency and to reduce latency in agent tool interactions.
• Tool comments: Adds rich metadata for the description of the tool behavior and enables an imaginative discovery and argumentation by AI agents.

Figure 1: Claude desktop with dramatic MCP for navigating and describing Datasette.io, which demonstrates the web automation operated by the model context protocol.

The protocol uses a modular JSON-RPC 2.0 basis, with a layered architecture of core transport, life cycle management, server functions (such as resources and input requests) and client functions (such as sampling or logging). Depending on the application, developers can select which components should be implemented.

The contribution by Microsoft: Browser automation via MCP

Two days agoMicrosoft published Dramatic MCPA server that integrates its powerful browser automation -tool in the MCP standard. This meansNteract with the web like real users.

The integration is based on the Tree Accomome accessibility and enables Claude to access and describe the side content in a human -readable form. The available tool set contains:

• navigation: browser_navigate, go_back, go_forward
• Entrance: browser_type, browser_click, browser_press_Key
• Snapshots: browser_snapshot, browser_screenshot
• Element -based interactions Use of accessibility descriptors

This transforms every compliant AI agent into a test automations -bot, QA assistant or data navigator.

People love MCP and we look forward to supporting our products.

Today in the agent SDK and support for the Chatgpt -Desktop -App + Answers API shortly!

– Sam Altman (@sama) March 26, 2025

Setup is simple: Users simply add dramatists as a command in Claude_desktop_config.json, and the Claude -Desktop -App recognizes the tools at runtime.

The overall picture: Interoperability in scale

Figure 2: With the modular design of MCP, developers can only implement the levels you need and at the same time maintain compatibility.

Anthropic first inserted MCP At the end of 2023 to solve a growing pain point: AI agents have to interact with actual tools, but each app speaks a different “language”. MCP aims to remedy this by providing a standard protocol for describing and using tools in all ecosystems.

With the support of anthropic, longchain and now Microsoft, MCP is a serious candidate to become the standard layer of the agent connection. Since MCP was first started by Anthropic, questions lingered whether Anthropic’s largest competitor Openai would support the protocol. And of course Microsoft, a great ally of Openai, was another question mark. The fact that both players have supported the protocol shows that the dynamics build up between corporate and open source communities. Openai itself opened its ecosystem for agents, including the latest agents that SDK announced a week ago – and The move has set the support in relation to Openais API formats to a standardIn view of the fact that others like anthropic and Google are in harmony. With the API formats from Openaai and MCP, the standardization has made a big win in the past few weeks.

“We enter the Ki protocol era”, ” tweeted Alexander Doria, co -founder of Ai Startup Pleias. “This is how agents will actually become Do Things.”

What’s next?

With the publication of MCP 0.2 and Microsoft’s tangible support, the foundations for a new generation of agents are laid who can think safely and flexibly over the stack and act flexibly.

Figure 3: OautH 2.1 Authorization flow in the model context protocol (MCP)

The big question now is: will others follow? If Meta, Amazon or Apple register, MCP could soon become the universal “language” of AI actions.

At the moment it is a big day for the agent ecosystem -one that brings the promise of AI interior operability closer to reality.