Anthropic has introduced a new protocol called “Model Context Protocol” (MCP) to create a standard that allows AI chatbots to connect with external data sources. This protocol enables chatbots to both read from and write to these data sources. In a demonstration, Anthropic showed how a chatbot generated an HTML page and directly saved it to a repository on GitHub. MCP is released as an open-source standard, intended to be accessible to all AI models.
The goal of MCP is to overcome the isolation of AI models, which typically interact with data systems only through custom interfaces. MCP is built on two main components: MCP servers and MCP clients. The servers provide data access and act as a bridge to various services. Anthropic has made MCP servers available for popular services like Slack, Google Maps, Fetch, Google Drive, and GitHub. MCP clients are the applications and workflows that access these servers.
A unique feature of MCP is its ability to access both local data and data on the internet. Initially, MCP can only be used on local servers, but this is expected to change soon.
For enterprise customers, MCP offers an incentive by making AI models more appealing for integration into business environments. Examples include embedding AI in development environments, connecting to internal company systems, and automating workflows. Anthropic has already partnered with some companies to start using MCP.
Developers can now access MCP, and Claude-Enterprise customers can connect their internal systems via MCP servers.
It remains uncertain whether Anthropic’s competitors will adopt MCP or prefer to develop their own solutions. For instance, OpenAI is working on a feature called “Work with apps,” which is currently available for ChatGPT on Mac. This feature allows users to read screen content from apps like the Xcode development environment, eliminating the need for copy-pasting to make specific requests. However, there is still no definitive information on the performance capabilities of MCP.