Until recently, an AI model could reason brilliantly about a problem but couldn't do much about it on its own. It could tell you what to check, draft what to send, describe what to build – but a person still had to open the tool, pull the data, and hand it back before the AI could take the next step. That's changed. AI agents can now connect directly to external tools and data sources, query them, and act on them – chaining several steps into a workflow they run themselves, with far less of a person relaying information between systems in the middle.
For research and media measurement teams, that shift is a big deal. The steps that used to add days to a project – checking feasibility, pulling fieldwork data, building out an analysis – can start happening inside the same conversation, at a fraction of the time. This is exactly the gap a new protocol is closing.
What actually is an MCP?
MCP stands for Model Context Protocol, an open standard released by Anthropic in November 2024. Before it existed, connecting an AI model to an external tool or data source meant building a custom, one-off integration for every single pairing – one for your CRM, another for your analytics tool, another for your research platform, each maintained separately. MCP replaces that patchwork with a single, common standard – one integration, not dozens.
It's often described as "USB-C for AI" – a universal connector that lets any AI model plug into any tool or data source without a bespoke integration built specifically for that pairing. An AI agent that supports MCP can, in principle, connect to any platform that offers an MCP server, and start querying or acting on it immediately.
From a research problem to an open standard, fast
The adoption curve has been unusually steep, even by AI standards. MCP server downloads went from roughly 100,000 in the month it launched to more than 8 million within six months, and the ecosystem now counts thousands of MCP servers connecting to everything from codebases to CRMs to research platforms. It hasn't stayed an Anthropic-only standard either – competing AI providers including OpenAI, Google and Microsoft have adopted it, which is part of why it has moved so quickly from a niche developer tool to something research and measurement platforms are now building for directly.
What this looks like inside a research team
In practice, an MCP connection to a research or measurement platform lets an AI agent do things that used to require a person moving between several tools: define an audience in plain language, check feasibility across markets, get pricing guidance, launch a study, monitor fieldwork as it comes in, and pull together a tailored analysis – inside the same conversation, without a manual export in sight. This isn't hypothetical. Cint, one of the larger players in the research and measurement space, announced an MCP collaboration with Potloc in September 2026 aimed at exactly this kind of AI-native workflow, and it's a strong signal that the rest of the industry will follow.
"An MCP doesn't just let an AI agent read your research data. It lets it act on it – launch the study, watch the fieldwork, and hand back a finished analysis, in the same conversation."
Why the data flowing through the pipe matters more, not less
MCPs solve a real bottleneck: the friction of getting an AI agent access to research data, and the ability to act on it. What they don't solve is whether that data is any good. An agent that can query and launch studies at machine speed doesn't fix a data quality problem – it accelerates whatever is already sitting in the pipe. Connect an MCP to a panel full of bots, synthetic respondents or gamed incentive-seekers, and the result is the same bad data, analysed and acted on faster than before.
That's exactly why voice verification at the point of entry becomes more important as AI agents get closer to the controls, not less. CHOOSI screens every respondent through AI voice technology before they enter a panel, so whatever queries that data next – a researcher reading a report today, or an AI agent pulling it through an MCP connection tomorrow – finds a real, authenticated human being behind every response, not statistical noise. And because CHOOSI's voice interviews already produce richer, more structured conversational data than a static survey export, that data is exactly the kind of nuanced context an agent needs to build a meaningful analysis, rather than skim a spreadsheet of top-line numbers.
MCPs are going to keep spreading through research and media measurement over the next few years, because the productivity case is obvious. The brands that get the most out of that shift will be the ones that made sure what's flowing through the pipe was trustworthy before they let an agent anywhere near the controls.
