Anthropic has released a new open source standard, the Model Context Protocol (MCP), which aims to solve the problem of isolating AI assistants from data sources and improve model response quality and relevance. MCP allows AI assistants to directly extract information from various data sources, avoiding "information islands" and thereby improving work efficiency. This technology establishes a two-way connection through an "MCP server" and an "MCP client". Developers can leverage this protocol to build AI-driven applications without the need to maintain separate connectors for each data source. Currently, several companies have integrated MCP into their systems, and Anthropic also provides pre-built MCP servers for common enterprise systems.
According to news on November 25, artificial intelligence company Anthropic announced the launch of a new open source standard - Model Context Protocol (MCP), which aims to improve the quality of model response to queries by connecting AI assistants with data sources such as business tools and software. and relevance. The release of MCP means that the AI assistant can directly extract information from different data sources when processing tasks, avoiding the problem of "information islands".
In a blog post, Anthropic said that although AI assistants have made rapid progress in inference and quality, most current models are still limited by being isolated from the data and unable to directly access the stored data. This requires a separate custom implementation for each data source, making the interconnected system difficult to scale. MCP aims to solve this problem with a protocol that allows developers to establish two-way connections between AI-driven applications (such as chatbots) and data sources.
The MCP protocol allows developers to share data through "MCP servers", build "MCP clients" (such as applications and workflows), and access these data sources through commands. Anthropic said developers can build using this standard protocol without having to maintain separate connectors for each data source, making the ecosystem more interconnected.
Currently, companies including Block and Apollo have integrated MCP into their systems, and development tool companies such as Replit, Codeium, and Sourcegraph are also adding MCP support to their platforms. Anthropic also said that subscribers to the Claude Enterprise plan can connect the Claude chatbot to their internal systems via MCP servers. In addition, Anthropic has shared pre-built MCP servers for enterprise systems such as Google Drive, Slack, and GitHub, and plans to launch toolkits to help enterprises deploy production MCP servers for the entire organization.
Although MCP has broad application prospects in theory, it is still unknown whether it can be widely supported, especially as competitors such as OpenAI are also launching similar functions. OpenAI recently introduced a data connectivity feature to its ChatGPT platform that allows AI to read code in coding-centric applications, a similar use case to MCP. However, OpenAI's approach is not open source, but implemented through close cooperation with partners.
At present, whether MCP can improve the performance of AI robots in tasks, as Anthropic claims, remains to be further verified.
Address: https://www.anthropic.com/news/model-context-protocol
All in all, the MCP protocol launched by Anthropic provides a new idea to solve the problem of isolation between AI assistants and data sources, and its open source characteristics also make it have wider application prospects. However, future development and market competition will determine whether MCP can eventually become an industry standard.