Microsoft is reportedly planning to adopt Anthropic’s Model Context Protocol (MCP) to let multiple artificial intelligence (AI) agents work together. As per the report, the Redmond-based tech giant said that it is focusing on multi-agent workflows and improved memory capability of its agents. The protocol standardises a way for AI chatbots to connect with external data hubs and retrieve information. The adoption of Anthropic’s MCP standard could be announced at the Microsoft Build 2025, scheduled for Monday. Notably, the Build conference is aimed at developers building apps and tools on Microsoft’s platform.
Microsoft Is Paving the Path to Create an Agentic Web
The tech giant is now planning to standardise how its AI chatbots and agents connect to external data hubs. Citing Microsoft’s Chief Technology Officer, Kevin Scott, Reuters reports that the company has a two-fold focus with its agents — to make multiple agents work together, and for them to develop a long-term memory.
Anthropic first introduced the MCP, a universal, open standard, in November 2024. The protocol aims to solve the problem of connecting AI systems with third-party data sources such as Google Cloud, Azure, and cloud-based customer relationship management (CRM) platforms.
Since each data hub operates differently, the backend process of connecting with these sources and retrieving data is also different. With AI players using their standard protocol, there is no uniformity in how the data is retrieved and processed, resulting in disparity in inference time as well as the final output. MCP provides a single protocol to let AI systems connect to these knowledge sources and access the data.
Scott reportedly said that Microsoft is focused on adopting standards across the technology industry to enable AI agents made by different developers to collaborate and work together. On MCP, Scott reportedly highlighted that the protocol can help the company create an “agentic web” similar to how hypertext protocols in the 1990s helped in spreading the Internet.
The Microsoft CTO reportedly also highlighted improving the memory function of AI agents to make them more useful. Most AI agents available publicly have enough memory to carry out the task assigned in a single prompt. While it can both provide comprehensive answers (Deep Research tools by OpenAI and Google), and execute tasks on the web (OpenAI’s Operator), it lacks the context from one task to another.
Microsoft is reportedly building a new approach dubbed structured retrieval augmentation. This system allows an AI agent to extract a small portion of information from each turn in a conversation with a user to gain the context of the series of tasks it performed in a session. It is expected that Microsoft will detail this approach, as well as how AI agents developed on its platform will follow MCP.
