Anthropic has released a new technical standard that can make it easier for AI agents to access apps on a user’s phone or desktop on their behalf. The new standard called Model Context Protocol (MCP) has been specifically developed to aid AI agent features like the recently rolled out ‘computer use’ capability of Claude 3.5 Sonnet. The open-source, universal standard is free to download by developers and can help with “connecting AI systems with data sources, replacing fragmented integrations with a single protocol.” “The result is a simpler, more reliable way to give AI systems access to the data they need,” Anthropic said in a blog post on Monday, November 25. As an example, the company said that information channels built on the MCP standard can help AI agents retrieve app data to better understand the context around a coding task. Currently, Anthropic has tied up with software companies Zed, Replit, Codeium, and Sourcegraph to adopt the MCP standard. With the MCP, developers do not need to depend on Google or any other AI company to create a custom connection between their AI agents and internal app data, according to a report by Gizmodo. Anthropic mentions that the two-way connections between AI systems and app data will be “secure” but does not provide further details on the potential privacy and security risks involved. Meanwhile, Apple and Google are also working on giving their own AI systems the ability to enter any app and directly take actions on it at the directions of the user.
