Search for “market research MCP” and most results are scraping, web-search and market intelligence data servers. They are good tools for a different job: they read what is already published. This guide covers the market research MCP servers that produce new evidence, by interviewing real people or by querying the interviews your team already ran, and compares them on what matters inside an agent client: who recruits, which modalities run, what evidence comes back, which clients connect and whether a person signs off before spend. For the product view, see the research API and MCP server for AI agents.
What can a market research MCP actually do?
Three jobs sit under the same label. Pick by job, not by directory category.
| Job | What the agent gets | Example servers |
|---|---|---|
| Run new qualitative research | A study designed from a brief, recruited participants, moderated interviews, findings with quotes | User Intuition, Outset, Listen Labs, Usercall, Koji, Voxpopme |
| Query research you already have | Answers grounded in past interviews, with links to source clips or quotes | Voxpopme, User Intuition, Listen Labs, Outset |
| Transcribe recordings you supply | Transcripts and exports for your own coding | Sonix, HappyScribe |
Market intelligence, web-scraping and search servers answer “what has been said publicly.” The servers above answer “what do the people I care about think, and why.” That second question is the one roadmap, messaging and pricing decisions turn on.
Which MCP servers can run new qualitative research?
Checked on 28 October 2026 against each vendor’s MCP documentation and public pages. “Not published” means those pages did not say.
| Vendor | Recruits for you | Modalities | Evidence returned |
|---|---|---|---|
| User Intuition | Vetted 4M+ consumer panel, B2B audiences recruited to order, or your own participants | Voice, chat, video | Findings with prevalence and verbatim quotes, each linked to its interview and recording moment |
| Outset | 25+ partner panels reaching 1.1B+ people in 85+ countries, per Outset; your own lists; or a shared link | Text, voice, video | Verbatims and highlight reels |
| Listen Labs | 50M+ network, or your own list | Text, voice, video, screen share | Analysis with citations linked to the moment in the recording |
| Usercall | 4M+ panel, or links and triggers | Voice, text | Themes with confidence scores and quotes |
| Voxpopme | Global panel, your list, or custom audience | Video, live interviews, AI voice | Answers linked to source clips |
| Koji | Managed panel, or share or embed a link | Voice, chat, phone | Per-interview insights, synthesized report |
| Vendor | Works in these agent clients | Approval before spend |
|---|---|---|
| User Intuition | Claude, ChatGPT, Claude Code, Codex, Cursor | Saved plan and cost estimate shown first; one click in Claude, a confirmation in the conversation elsewhere |
| Outset | Claude, ChatGPT, Cursor, plus other MCP-compatible tools | Not published |
| Listen Labs | Claude, ChatGPT, Codex, Cursor | Governed by workspace launch permissions; no per-launch review step published |
| Usercall | Claude, ChatGPT, Codex, Cursor | The agent has no share or activate step; a person shares the interview link |
| Voxpopme | Claude, ChatGPT, any MCP client | Not published |
| Koji | Claude, Claude Code, Cursor, VS Code, Windsurf | Not published |
Sources, retrieved 28 October 2026: Outset MCP, Outset recruiting, Listen Labs MCP connection docs, Listen Labs MCP announcement, Listen Labs evidence links, Usercall MCP and Agent API docs, Usercall panel, Voxpopme MCP announcement, Koji MCP integration, Koji docs. REST access, webhooks, authentication and published pricing for the same vendors sit in the best research APIs comparison, and the agent workflow head-to-head adds Quals.ai.
Two things the tables cannot show decide most evaluations. The first is depth: User Intuition’s AI moderator runs 20–30 minute interviews that ladder 5–7 levels deep, following each answer down to the motivation underneath it. The second is trust: 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.
What does a finished study look like?
Inside the agent, a research MCP turns a brief into a saved discussion guide and screener, asks you for anything the plan still needs, and reads the findings back when fieldwork ends. 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. The MCP research API guide covers each stage and the tools behind it.
What comes back is the part worth judging. The example below comes from a User Intuition study of 43 Walmart shoppers, each reconstructing one recent trip in a voice interview. 34 of 43 hit at least one obstacle on that trip, and 19 of 43 named the shelf as the moment that shaped the trip. Each count names the interviews behind it, so a reader can open the conversations that produced it. Above the counts sits the study’s headline, the one line a stakeholder remembers:
“This is not loyalty. It is tolerance, and it is priced.”
You can explore the Walmart study yourself: read the findings, browse the transcripts and listen to sample interviews.
What should I tell the agent to get a good study?
The brief does most of the work. A research-platform MCP turns a clear brief into a discussion guide, a screener and a recruiting plan, so give the agent the same four things you would give a senior researcher:
- The decision. “We are choosing between two packaging concepts for the spring launch,” not “learn about packaging.”
- The audience. Who counts, who doesn’t, and the country. “US shoppers who bought a frozen dinner in the last two weeks, mixed age and income.”
- What you already believe. Hypotheses the study should test or break, so the moderator probes them instead of circling them.
- The output you need. “Which concept wins and why, with the strongest quote for each side,” or “the three reasons shoppers switched brands, ranked by how often they came up.”
A prompt that works in any connected client:
Run a voice study with 20 US shoppers who bought a frozen dinner in the last two weeks. We need to know what made them pick one brand off the shelf over the others, and what nearly stopped them. Show me the plan and the cost before launching.
The agent creates the study, asks you anything the plan still needs, and comes back with the guide and the estimate for your sign-off. A vague brief gets a vague guide back, so if the first plan reads generic, the fix is usually a sharper decision line, not a longer prompt.
How do I choose between them?
- You have no participants of your own: all six recruit, so compare panel quality rather than panel size. User Intuition’s vetted 4M+ consumer panel is checked on video and tracked across studies, so a participant can’t claim one profile this week and another next week, and B2B audiences are recruited to order per study.
- You need depth on why, not just what: compare interview length and follow-up behavior. Ask each vendor how far the moderator follows a vague answer before moving on; User Intuition’s moderator adapts every follow-up to the last reply.
- You need a person to sign off before spend: check the approval column. Only some vendors publish a step where a person reviews the plan and the cost before fieldwork starts.
- Your team works in a specific client: check the client column. A server documented for Claude and ChatGPT may say nothing about Codex or Cursor, and the reverse.
- You already hold a large research repository: Voxpopme, Listen Labs and User Intuition all let an agent ask questions across past studies with cited answers; see how to ask questions across past research.
- You need video or a prototype test: check modality. User Intuition runs voice, chat and video, plus concept tests with images and prototype tests with a live link.
- You need a price your agent can quote: User Intuition publishes $30 per voice interview, and the research APIs comparison shows which other vendors publish theirs.
- You need a directional read before fieldwork: User Intuition’s synthetic respondents are also available in chat for a directional first read.
How does an agent show where each finding came from?
Every User Intuition finding carries its evidence with it. Prevalence says how many participants raised the point, so a finding shared by most of a sample reads differently from one raised by three people. Each supporting quote is verbatim and links to its source interview, and in voice and video studies to the moment in the recording, so a reader can hear the participant say it. When the agent writes a summary or a recommendation, it keeps those links in place, which means the decision-maker can check any claim by opening its source instead of trusting the summary. The agent’s output stays grounded in what real people said rather than in what a language model assumed about them. That matters most when a finding is surprising: the first thing a skeptical stakeholder asks is who said it, and the answer is one click away.
The same evidence carries into follow-up work, so the agent can turn a study into a presentation or a brief for the team without leaving the conversation.
How do I connect a research MCP to ChatGPT or Claude?
In ChatGPT, install the User Intuition plugin from the plugin directory in one click, sign in, and describe what you need to learn. ChatGPT drafts the study, shows the saved plan and estimate, and launches after you confirm in the conversation. In Claude, add User Intuition Research from the connector directory; User Intuition’s skills for Claude guide each stage, from brief to findings, and you approve a launch in one click. In both, the assistant can answer follow-up questions across your past studies, with every quote linked to its interview. Codex, Cursor and other MCP clients connect with the same server (setup). To try it, create an account: signup includes 3 free interviews with your own participants; panel recruiting is billed separately.