Anthropic released a new, unified open-source method to connect models with data

Anthropic has shared in a blog post that it has developed a new standard to connect AI-powered assistants with data sources (filesystems, apps, repositories) known as the Model Context Protocol (MCP). MCP works with any model, not just Claude, and resolves a long-standing challenge when deploying AI-powered solutions: isolation from data sources. The startup announced it is open-sourcing MCP so the community can take advantage of its benefits and continue building upon it.

In addition to addressing the problem of configuring multiple sources individually, Anthropic claims that connecting models and data with MCP allows models to produce better, more relevant outputs. The company also reports that early adopters including Block and Apollo are already implementing MCP within their systems, while other companies, such as Zed, Replit, Codeium, and Sourcegraph, are using MCP to improve their platforms by, for example, enabling their coding assistants to retrieve more relevant context to a coding task, which translates into "more nuanced and functional code with fewer attempts."

Currently, the MCP has three main components: the Model Context Protocol specification and SDKs, MCP server support in the Claude desktop, and a repository of pre-built MCP servers to connect AI systems with business tools including Google Drive, Slack, GitHub, Git, Postgres, and Puppeteer. These components allow developers to start building custom local MCP servers tailored for their systems and contribute to the open-source repository; and enable Claude for Work customers to install pre-built MCP servers to connect Claude to their internal tools and datasets. Additionally, Anthropic plans to release tools for deploying MCP servers that serve entire organizations.