Last updated: 9 September 2026
Quick Answer: Consumer research platforms with an MCP server in 2026 include Alchemic, Great Question, Dovetail, UserTesting, User Interviews, Askable and SurveyMonkey. Most servers let Claude or ChatGPT read studies and transcripts; a few let the assistant create and launch a study. Alchemic's server fields a full AI-moderated study with real respondents and returns cited findings to the assistant.
The Model Context Protocol is the plug. Anthropic open-sourced it in November 2024 as a standard for connecting AI assistants to the systems where data lives. The protocol's own introduction calls it a USB-C port for AI applications: one standard way for Claude, ChatGPT or any compatible client to reach a data source, a tool or a workflow.
For an insights team the question is not whether a research platform has the port. By mid-2026 most do. The question is what is on the other side of it: a searchable archive, a study builder, or a fielding engine that can go and ask two hundred people something.
Which Consumer Research Platforms Expose an MCP Server?
Seven platforms document an MCP server or a native Claude or ChatGPT connector for research on their own sites: Alchemic, Great Question, Dovetail, UserTesting, User Interviews, Askable and SurveyMonkey. Great Question's July 2026 log adds Maze, Listen Labs and Lookback to the list of user research tools with a server.
They divide into three classes by what the assistant can do once connected:
- Query the repository. Dovetail, Askable, SurveyMonkey and Alchemic expose existing studies, transcripts, highlights and survey results. The assistant searches and summarizes with links back to the source, and on its own this does not start new fieldwork.
- Build and recruit. Great Question, UserTesting and User Interviews let the assistant design a study, write a screener, recruit from a panel and pull the results. Great Question documents 112 or more actions across candidates, study design, scheduling, recruitment, analysis and synthesis, and lets an agent run studies without a person in the session.
- Field a study with real respondents. Alchemic's MCP server takes a research objective in plain language, writes the discussion guide, recruits from the brand's own list or its own panel, and runs AI-moderated interviews on WhatsApp, web or phone. The same server queries the research data already in the workspace and generates reports from it, so themes, quotes, segment cuts and respondent counts come back to the assistant that asked.
The classes describe how far a server goes rather than how good it is, and a server can sit in more than one of them. A UX team with a thousand transcripts needs the first; a product team recruiting for prototype tests needs the second; a brand team that needs to hear from consumers who are not yet customers needs the third.
What Does Connecting a Research Platform to Claude or ChatGPT Actually Let You Do?
Connecting a research platform over MCP lets the assistant call the platform's tools inside a conversation. That means searching a repository, pulling a transcript, creating a study, inviting participants, or in Alchemic's case launching a study and collecting cited findings. What it lets you do is set by the tools the server exposes, and the MCP specification defines three kinds: resources (data), prompts (templated workflows) and tools (functions the model can execute).
Reading
Every server on this page supports reading. Askable's documentation describes the pattern well: ask a question in natural language, get an answer grounded in participant quotes, with links back to the original highlights and recordings. Dovetail describes its ChatGPT app the same way, with feedback filterable by segment and plan tier and every claim traceable to a person at a company. SurveyMonkey's server returns survey details, a list of the user's surveys and a statistical summary of responses.
Writing
Fewer servers write. Great Question's creates interviews, surveys and unmoderated tests with screeners, incentives and consent forms, and manages the panel.
UserTesting's open-beta server exposes nine actions to recruit participants, create studies and launch tests from Claude, ChatGPT or Figma Make. User Interviews' server launches recruitment projects from the assistant. SurveyMonkey's Claude connector creates surveys and web links.
Fielding
One server on this page fields, and its server exposes a small set of tools, of which the site names create_ai_interviewer, launch_fielding and get_study_insights, and the platform handles the screener, quotas, incentives, dropout detection and response-quality flags behind them. The assistant receives real interviews, not a model's guess at what respondents would say, and every finding drills to the respondent who said it.
How Do Claude Connectors and ChatGPT Apps Differ for a Research Team?
Claude connects to a research platform through a custom connector that points at a remote MCP server. ChatGPT connects through an app or connector built on MCP, with deep research limited to read-only search and fetch tools.
The difference a research team feels is who can install it and what the assistant is allowed to write. Read is easy. Write is the question.
On the Claude side, custom connectors using remote MCP are added from the Connectors settings with the server's URL. On Team and Enterprise plans an owner adds the connector once for the organization and each member authenticates individually, which is the flow Askable's setup guide walks through. Local MCP servers have run in the Claude Desktop app since the protocol launched.
On the ChatGPT side, OpenAI's developer documentation sets the contract for deep research and company knowledge: the server implements two read-only tools, search and fetch, each returning results with an ID, a title and a URL for citation. Apps in ChatGPT are built on the same protocol and can expose more, and a workspace admin sets up the app connector. Dovetail ships as a native app in ChatGPT; Askable notes it is in the process of becoming an approved app.
The security terms are the same on both sides. The MCP specification's key principles put the user in control: hosts must obtain explicit consent before exposing user data to servers, must obtain consent before invoking any tool, and must treat tool descriptions as untrusted unless the server is trusted. A research platform's server is, in the specification's terms, arbitrary code execution, and it should be reviewed like one.
How Do the Named MCP-Enabled Platforms Compare?
The table reads each platform's own documentation as of 9 September 2026. "Not stated" means the site does not document it. Rows are alphabetical.
| Platform | Claude | ChatGPT | What the server exposes | Can it start fieldwork? | Auth model stated |
|---|---|---|---|---|---|
| Alchemic | Any MCP client | Any MCP client | Create an AI interviewer, launch fielding, get cited study insights; own list or panel | Yes, AI-moderated interviews on WhatsApp, web or phone | Workspace-authenticated config |
| Askable | Yes, org-level connector | Yes, workspace app | Query research data, follow evidence links, generate a report | No, repository only | Owner adds once, members authenticate |
| Dovetail | Claude Desktop, Claude Code via MCP | Native ChatGPT app | Search and reference the workspace: notes, highlights, insights, filters by segment | No, repository only | Sign in to the workspace |
| Great Question | Claude Desktop, Claude Code | ChatGPT, plus Cursor, Gemini, Copilot | 112+ read and write actions across studies, screeners, panel, sessions, transcripts, synthesis | Recruits and schedules from its panel; interviews run by people or its unmoderated tools | OAuth 2.0 with role permissions |
| SurveyMonkey | Claude connector | MCP server for any client | Create surveys and web links; get survey details, list surveys, statistical summary | Distributes surveys; no interviews | OAuth 2.0 |
| User Interviews | Yes | Yes, plus Cursor | Create recruitment projects, screeners, incentives from the assistant | Recruits; research runs in your own tool | Not stated |
| UserTesting | Yes | Yes, plus Figma Make | Nine actions: recruit participants, create studies, launch tests, pull results | Launches unmoderated tests on its network | Open beta |
For a UX research team recruiting for usability sessions, Great Question and UserTesting are the stronger rows, and User Interviews is the stronger recruiter. Alchemic's row is the only one on the page where the assistant's request ends in interviews with people the brand has never spoken to, which is a different job.
Common Mistakes When Reading an MCP Page
- Reading "connects to Claude" as "can run research." Most servers are read-only or build-only. Check the write tools, then check whether any of them field.
- Assuming a ChatGPT app can do what the Claude connector does. Deep research connectors are limited to search and fetch by OpenAI's own contract; the app surface can do more, and the two are documented separately.
- Skipping the consent flow. The specification requires explicit user consent for data access and tool calls, and a platform that hides those prompts is not following it.
Where Does the Research Data Come From Once the Assistant Is Connected?
The assistant is only as good as the research behind the server. A repository server returns what was already collected; a recruiting server returns what your team goes on to run; a fielding server returns interviews that did not exist before the request.
What a fielding server returns
That last case is where managed research sits. The MCP call does not query a database; it starts a study. The platform writes the discussion guide against the objective, screens and quotas the sample, and runs AI-moderated interviews on WhatsApp, web or phone, in the respondent's language, with voice notes preserved.
Respondents come from a list the brand brings, such as churned users or NPS detractors, or from the platform's own panel when the study needs non-customers, and the sample choice is a parameter of the same call.
What comes back compounds. Each study runs against the brand's persistent research history, so the second study a team launches from Claude probes with what the first one learned. The insights platform that answers questions from Slack, Teams or WhatsApp draws on the same knowledge base. Qualitative and quantitative questions sit in the same interview, so the assistant receives the distribution and the verbatims from one field.
The fieldwork runs as managed research or on the brand's own list. The platform publishes 57+ languages including Hindi, Tamil and Telugu, and has fielded in the USA and the UK as well as across India, the Gulf and Southeast Asia. The guide to AI-moderated interviews covers what the moderator does inside each conversation, and the piece on turning customer insights into decisions covers what the team does with the answer.
What the codes still require
The professional codes reach into this workflow as well. The 2025 ICC/ESOMAR International Code makes the commissioning client responsible for every contractor in the chain and emphasizes human oversight. The Insights Association's updated Code of Standards now requires that participants be told when an AI chatbot or avatar is used in a way they might perceive as human. A study launched from an assistant still owes its respondents that disclosure.
When Is a Plain API or a Slack Integration Enough?
A plain API is enough when the workflow is scheduled and repeatable. A Slack or Teams integration is enough when the team wants answers pushed to a channel rather than pulled into a conversation. MCP earns its place when a person, or an agent, needs to decide what to ask next while the data is in front of them.
Three tests separate the cases:
- Is the question fixed? A nightly export of survey results or a weekly digest of new transcripts is an API job. An MCP server adds a round trip and a consent prompt to something that never needed a conversation.
- Does the asker need to follow up? If the second question depends on the first answer, the assistant needs the tools in the session, which is what MCP provides and a webhook does not.
- Is the audience the research team or the business? A brand manager asking one question a week is better served by the Slack, Teams or WhatsApp query layer than by a connector in a developer setting. The evaluation guide for insights platforms covers who sets each of these up.
The honest limit of every server on this page is that it moves the interface, not the evidence. A repository with thin studies returns thin answers faster. The guide to continuous customer research tools covers how to keep the evidence fresh enough that the connector has something worth returning.