What is it?

The model context protocol (MCP) is an open standard for connecting AI applications and large language models (LLMs) to external data, tools and systems. MCP was created by AI firm Anthropic, the developer of Claude, in 2024 and is designed to allow similar services to access information and perform tasks based on the external systems and data provided.

Anthropic describes MCP as akin to a USB-C port, allowing other organisations to make their tools, data or workflows, such as specialised prompts, available to the AI application. Once connected, users can use the AI model to analyse or carry out tasks using the external information provided.

MCP does not, by itself, train the underlying model on the data connected, but allows the AI application to use it for relevant context or to invoke tools, subject to the data handling policies of the systems involved.

Not to be confused with: API

An application programming interface (API) is a general-purpose interface between two software systems, or simply, a set of rules that allows software applications to communicate and share data. APIs let one service request information or carry out specific actions in another without being part of that service.

MCP is a common protocol for giving AI applications powered by LLMs a standard way to use tools, data or systems. MCP servers may use APIs to connect to external services; however, MCP relates specifically to AI applications and provides a more standardised way to allow them to access external information, tools and systems.

Why does this matter for the research sector?

For organisations with large volumes of data based on original research, there is a real benefit to allowing LLMs to help customers analyse and work with their datasets and systems. Tools such as Claude and ChatGPT are widely used, and these familiar interfaces can help increase the accessibility of research data.

Several research companies have announced MCP compatibility. The benefits of installing a bridge to connect research platforms to people’s preferred AI tools includes the elimination of manual file uploads, real-time data access and cross-platform synthesis. There is the potential for users to get instant answers to queries, as well as accessing in-depth analysis.

Ray Poynter, chair of Esomar’s professional standards committee, said: “MCP is a protocol that enables AI systems such as ChatGPT and Claude to connect with, use and control other software, for example Canva. In the short term, this can improve the quality and usefulness of outputs: an AI could help process data, then produce a well-structured report or presentation directly in Canva.

“MCP may not itself be the whole story. We are also seeing the growth of reusable AI skills that can work across platforms, and more ways for AI to operate software directly even where an MCP connection does not exist. The larger shift is towards AI driving outputs from the programmes we use.

“For researchers, that means linking AI more directly to survey tools, project management, analysis and visualisation, with AI increasingly becoming the operating system for market research work.”

• The full series can be found here.