MCP for market research is the use of the Model Context Protocol (MCP) to let an AI assistant such as Claude, ChatGPT or Cursor call a research platform’s tools. The protocol is plumbing. What it changes for researchers is that the assistant you already draft in can commission a study with real people, not just summarize a study someone ran last year.
This guide explains what MCP is, how its parts map to research work, what it means for qualitative research in particular, how read-only and read-write research servers differ, and who uses it. It is a definition, not a setup manual. When you are ready, you can connect it in Claude, ChatGPT, Codex or Cursor in a few minutes (setup).
What is MCP (Model Context Protocol)?
MCP is an open protocol that defines how an AI application talks to an outside tool server. Anthropic introduced it as an open standard in November 2024, and OpenAI, Google, Microsoft and hundreds of tool vendors have since adopted it. The protocol is open source, so any assistant or platform can implement it.
Three roles make up an MCP connection. The host is the application you work in, such as Claude Desktop, ChatGPT or Cursor. The client is the part of that application that holds the connection. The server is the outside system, here a research platform, that advertises what it can do. When the client connects, the server returns a catalog of capabilities with typed inputs, and the assistant decides when to call them.
Before MCP, every pairing of assistant and tool needed its own integration: one for ChatGPT and a research tool, another for Claude and the same tool, and so on. MCP collapses that into one integration per server. A useful comparison is USB. One port standard replaced a drawer of proprietary cables, and one protocol now lets a research server built once work in every compatible assistant.
For a market researcher, the practical meaning is short. If a research platform publishes an MCP server, the assistant you already use can operate that platform on your behalf, in the same conversation where the question came up. You do not learn a new interface, and the assistant does not need a custom plugin for each vendor.
What is MCP for qualitative research?
MCP for qualitative research means the assistant a researcher already uses can commission interviews with real people and read the results, with no separate tool and no code. The researcher writes a brief in plain language, the assistant turns it into a study through the research server, and the findings come back into the same conversation with each point tied to verbatim quotes.
Imagine you are working in Claude on a packaging brief and want to know how shoppers react to two front-of-pack lines. Before MCP, you would pause, open a research platform, configure a study, wait for results, and paste the findings back. With MCP, you ask: “Ask 15 US snack buyers which of these two pack lines makes them want to try the product, ‘Baked, not fried’ or ‘Real cheese, real crunch’, and why.” The assistant plans the study, shows you the saved plan and estimate, and after you approve brings the findings back into the conversation. You never left the brief.
What MCP does not change is the method. A qualitative interview is still a moderated conversation that follows each answer down to the reason behind it, and the quality of that conversation depends on the research platform, not the protocol. The researcher still owns the decision the study serves, the audience, and the stimulus. What moves into the assistant is the operational work around the interview: drafting the study, relaying the platform’s planning questions, tracking fieldwork, and reading the report. Because findings return as structured data with quotes attached, the assistant can reason over them, compare them with earlier studies, and draft the next brief without losing the link to what participants said. The evidence also outlives the conversation: each study joins the account’s research library, so the next question can start from what earlier interviews found.
What do tools, resources and prompts mean in research terms?
An MCP server exposes three kinds of capability. In a research server, each maps to a familiar part of the research job.
| MCP primitive | What it is | In a research server |
|---|---|---|
| Tools | Callable operations with input schemas | Create a study, run the planning step, estimate a panel, launch fieldwork, list interviews, read each transcript, retrieve a report, ask across past research |
| Resources | Readable context the agent can load | Supported languages, panel countries, study types and interview modes |
| Prompts | Reusable templates for a workflow | Step-by-step guidance for briefing, fielding and analysis |
Tools carry the weight. In User Intuition’s server they include create_study, customize_study, estimate_panel, launch_panel, list_interviews, get_interview and get_study_report. Each has a single clear effect, so reading a report is a separate call from launching paid recruitment, and a launch waits for a person’s approval.
Resources keep the agent honest about what is possible. Before proposing a study in Brazil in Portuguese, the agent can read which countries and languages the account can field, rather than guessing.
Prompts encode good order. Research has a natural sequence, from brief to plan to approval to fieldwork to analysis, and prompts help the agent follow it.
Discovery ties the three together. The client asks the server what it currently exposes, so the agent always works from the live catalog. The tool reference lists the full set.
What do connectors, MCP Apps and skills mean?
Researchers meet a few product terms on top of the protocol itself. Each one names a layer around the same server.
- Connector. The name Claude and ChatGPT give an MCP server you add to the app. For most researchers, adding the connector is the whole setup, and the research tools then appear in the conversation.
- MCP App. An extension of the protocol that lets a server show interactive elements inside the conversation, such as a card for reviewing a plan. Claude supports MCP Apps, which is why approving a research plan and estimate there is one click; in other clients the agent shows the plan and estimate and you confirm in the conversation.
- Skills and plugins. Packaged instructions that teach an agent a workflow on top of a server’s tools. In Claude Code, the
user-intuition-researchplugin includes the hosted server and adds 10 research skills, listed in the skills library. - Tool call. One use of one tool. In the conversation it shows up as a step, such as creating the study or fetching the report, so you can see what the agent did on your behalf.
What is the difference between a hosted and a local MCP server?
MCP servers run in one of two places, and the choice decides how you sign in and what you install.
A hosted server, also called a remote server, runs on the vendor’s infrastructure and is reached over the internet. You add its address to your assistant and sign in with OAuth, so nothing is installed on your computer and access is tied to your account. This is the usual path for researchers working in Claude or ChatGPT.
A local server runs as a process on your own machine, started by the assistant over standard input and output. It usually authenticates with an API key held in an environment variable. Local servers suit developers working in an IDE, or scripts that should run without a browser sign-in.
User Intuition offers both. The hosted server lives at https://mcp.userintuition.ai/mcp, and the local package is @userintuition-ai/mcp. Both reach the same account, the same studies and the same research library, so a study started from one can be read from the other.
Why does MCP matter for market research?
Most AI tools that touch research operate in read-only mode. They summarize transcripts, tag themes in uploaded files, and search across past studies. Those jobs are useful, and they share a ceiling: if nobody has studied a topic, the AI has nothing to read. A read-write research server removes the ceiling, because the agent can create knowledge that did not exist before the question was asked.
The second shift is distribution. Questions that need evidence from people come up in strategy documents, campaign drafts and product reviews, while the research tool lives behind a separate dashboard. That gap is why most questions never become studies. With MCP, the question and the evidence sit in the same conversation. For more on why grounded evidence matters to agents, see why AI agents need real consumer data.
What is the difference between read-only and read-write research servers?
The distinction decides what an assistant can do when your existing research has no answer.
| Capability | Read-only research tools | Read-write research MCP |
|---|---|---|
| Search past research | Yes | Yes |
| Summarize transcripts | Yes | Yes |
| Tag and theme existing data | Yes | Yes |
| Design a new study from a brief | No | Yes |
| Recruit real participants | No | Yes |
| Run moderated interviews | No | Yes |
| Return new findings with source quotes | No | Yes |
A read-only tool works like a research librarian, organizing what you already hold. A read-write server works like a research team on call. When a brand manager asks what shoppers think about a new pack size, a read-only tool searches old pack-size findings. A read-write connection runs a new study with real shoppers and returns fresh evidence, and most panel studies fill within 24 hours.
Write access also brings a spending decision, so a well-built read-write server separates reading from launching. The agent shows you the saved plan and a cost estimate before anything is spent; in Claude you approve with one click, elsewhere you confirm in the conversation.
What does running a study over MCP look like?
In short, the agent turns a plain-language question into a draft study, relays the platform’s planning questions, shows the saved plan and estimate for approval, and returns findings tied to quotes once interviews finish. Each step and what it returns is documented in the MCP research API reference.
Worked runs live with the workflow they serve: how to run a concept test from an AI agent for packaging and product concepts, and the consumer research API guide for claim tests and brand-switching studies.
Who is MCP for in market research?
Every client connects to the same server and runs the same studies. What differs is who is asking and where the work happens.
Insights and brand teams. Researchers and brand managers who already draft in Claude or ChatGPT commission a focused study mid-draft, read the findings, and ask follow-up questions without switching tools. The agent handles setup, and the researcher reviews the plan and the evidence. Researcher time shifts toward the larger programs where study design judgment matters most.
Product teams in coding tools. Developers and product managers who spend the day in Cursor, Codex or Claude Code connect the research server once and ask user questions beside the code: whether a checkout step reads clearly, or whether a feature name makes sense to buyers. See agentic research for product teams.
Teams building research agents. Teams with engineering resources build pipelines on the same tools, such as an agent that runs a reaction study when a competitor launches a new pack, or one that briefs a follow-up when satisfaction drops. MCP gives that agent a ready integration surface.
Founders and small teams. Without a research department, the founder’s assistant becomes the research operator. It plans, recruits and returns cited findings while the founder makes the call.
How does MCP research handle privacy, consent and export?
MCP-driven research runs under the same terms as research started from the dashboard, because both reach the same account. User Intuition complies with the EU GDPR and CCPA/CPRA, and does not use customer data to train, fine-tune or improve AI models.
- Scoped access. Hosted connections sign in with OAuth and local connections use an API key, and either one reaches only your account’s research.
- Consent. The platform, not the agent, collects participant consent before each interview.
- Storage. Transcripts and recordings stay with the study, where your team can review them.
- Export. Interviews export as CSV or JSONL files, with common personal identifiers masked by default, so an analysis pipeline receives the content without the contact details.
The security page covers the full data-handling posture.
Why does the research server matter more than the protocol?
MCP standardizes the connection, and it says nothing about the quality of the interviews on the other side. Two servers can expose a tool called “create study” and return very different evidence, which is why choosing a market research MCP server comes down to depth and trust.
User Intuition’s AI moderator runs 20–30 minute interviews that ladder 5–7 levels deep, following each answer from what a shopper did to how it felt, in voice, chat or video and in 80+ languages. The people on the other end are a vetted 4M+ consumer panel, with B2B audiences recruited to order per study, or your own customers. Every panelist call is reviewed by the User Intuition team, panelists pass a video identity check, and only interviews that pass quality scoring are billed; see how quality works. User Intuition holds 4.9/5 on G2 and 5/5 on Capterra, with 98% participant satisfaction, and each finding carries its prevalence and the verbatim quotes behind it. To see that structure in a finished report, explore an example study.
How do you start with MCP for market research?
Create a free account; signup includes 3 free interviews with your own participants, and panel recruiting is billed separately. Connect it in Claude, ChatGPT, Codex or Cursor in a few minutes (setup), then run one small study on a real decision and ask the agent to show you the plan and estimate first.
MCP is the connection, and agentic research is what an agent does with it; what is agentic research covers the practice end to end. For the product view, see the research API and MCP server for AI agents.