The ultimate guide to UX research with MCP
The guide to running UX research from inside Claude, Cursor, or ChatGPT. Written for the people approving it, the people rolling it out, and the people actually using it.
The Great Question MCP server connects your research repository to your AI client of choice. Every study, every transcript, every highlight, every insight. Once connected, your AI can read your research, design new studies, recruit panelists, and synthesise findings without leaving the chat.
Included free on Pro and above. No charge per tool call. We don’t meter MCP usage.
Customer research is moving inside the AI client. The MCP is how Great Question gets there with you.
The guide
Who this is for
Pick your seat. This guide is structured top to bottom in the order most enterprises read it.
Security, IT, ReOps, vendor risk. The "if I flip it on, what happens?" reader.
Head of Research, Product, Design, CIO. The "how do I do this in waves?" reader.
Researcher, PM, designer, CSM, exec. The "what can I actually do?" reader.
Same nuance as above. The evidence you want is in customer stories.
Quick start
Install it yourself. No engineer. No API key.
No procurement gate. Pick your client, paste the server URL, authorize in the browser. First prompt in under 5 minutes.
- Profile picture → Customize → Connectors → Add custom connector
- Name: Great Question. URL: paste the server URL. Click Connect.
- Authorize in the browser window that opens.
- Restart Claude Desktop.
- Try: “What studies have I run in the last 6 months?”
claude mcp add --transport http GreatQuestion https://greatquestion.co/api/mcp/v1
Approve in the browser. Verify with claude mcp list. Done.
Open settings (Cmd-Shift-J). MCP tab → New MCP Server. Paste:
{"mcpServers":{"great-question":{"url":"https://greatquestion.co/api/mcp/v1"}}}Save, Connect, authorize.
Settings → Apps → Advanced settings → toggle Developer mode. Create app with name Great Question, paste server URL, OAuth. Authenticate via browser.
Windsurf, Zed, VS Code, Codex, Jules, Cline, Gemini, Cowork — standard MCP HTTP transport with the server URL above. Standard OAuth 2.0. No API keys.
What you can do
15 real prompts
Prompts our customers actually paste into Claude regularly.
Customers, on record
Real names. Real quotes.
From product calls and live demos in the last 30 days.
“Building our own research stack makes no sense.”
“This feels like a long-time vision coming together. The value comes from the connection of all of these data sources together — that’s huge.”
“It can 10x our output.”
“The time to plan, conduct, and analyze research is the biggest bottleneck. If we can integrate automated workflows, the potential for research to be more impactful is so much more evident.”
“Are you in Claude all day now? Basically, yes.”
Jose Rodriguez, lone researcher at Confluent. Enabled MCP live on the call.
“No prototype, no meeting.”
Built a Claude skill on top of Great Question MCP that creates synthetic users from past interview transcripts and pre-validates designs against them.
“The MCP is our wedge.”
“UI will shrink, CLI will grow.” DeepL is deprecating UserTesting specifically because it’s not AI-native.
Running a pilot on MCP + repo
Paying to integrate research with Claude for “speeding up the creation of shaping docs and PRDs.”
“For us internally, we basically don’t work with any vendors anymore that don’t have an MCP and aren’t leaning aggressively into AI.”
Security
What happens when you turn MCP on?
The most common enterprise question: “If I flip it on, can my entire org suddenly see everything?”
[ AI Client ] [ Great Question ] [ Your LLM ]
Claude/Cursor MCP server Anthropic/
ChatGPT/Code api/mcp/v1 OpenAI/etc.
| | |
| 1. OAuth 2.0 | |
|-------------------> | |
| 2. Tool call | |
|-------------------> | |
| RBAC + PII redaction |
| 3. Result | |
|<------------------- | |
| 4. Client -> LLM | |
|----------------------------------------->|What's new
Updated weeklyCursor tool-name length bug fixed. Cursor users now see the full tool set.
Listed on Smithery, LobeHub, Glama, PulseMCP, MCP.so, Cline, and the GitHub MCP Registry.
Public GitHub repo for the MCP server (GreatQuestion/mcp-server).
Write operations GA. Create studies, screeners, candidates, send invites. All from your AI client.
OAuth 2.0 replaces API keys. Workspace admins enable MCP from AI & MCP settings.
What's coming next
MCP v1.3
IN FLIGHTSession and transcript-title PII redaction, prototype-test aggregate analytics, incentive fields on study create/update.
Auto-generated tool reference
Continuously-updated tool documentation at /mcp/tools, regenerated on every deploy.
MCP-led onboarding
New accounts go from install to 3 ready-to-run studies in under 10 minutes with gq_bootstrap_account.
MCP Intelligence Tools
Orchestration above CRUD: gq.insights, gq.status, gq.connect, gq.recruit.
Enterprise controls
Per-user enable/disable, AI-client domain allowlist, audit-log export.
Sandbox / trial environment
So you can try MCP before committing. In exploration with engineering.
Rolling out MCP across your org
Customers who get the most out of MCP roll it out in waves rather than blasting it org-wide on day one.
Research + ReOps
Senior researchers and your ReOps lead. Prove value, build the prompts library, set the tone for governance.
Product + Design ICs
PMs and designers who consume research the most. Shift research consumption left. Reduce find-me-a-quote requests.
CS, Sales, Support, Exec
Customer-facing teams and leadership. Spread customer voice deeper into the org.
Governance and quality
The biggest risk with MCP is not security. It is sloppy AI-generated synthesis that looks authoritative but is not.
Where they have used AI in terms of design is to do synthesis, but really sloppy. And so sloppy outputs leads to sloppy design, which is, yes, terrible outcomes.
How do you think about automating the QA process? The report it automatically generated, the quality is just not quite there yet. So we still have to spend a lot more time to review the data, see if AI hallucinated.
ReOps becomes the enforcer. Tag the repo well, author rubric prompts, run AI outputs through a human review loop. Most customers use MCP for synthesis and still bring humans in for the final framing.
Frequently asked
The questions we hear on every prospect and security call.
If I turn it on, can my whole org access everything?▾
No. MCP inherits your existing Great Question RBAC. A user can only see and do via MCP what they can already see and do in the UI.
Is my data used to train AI models?▾
No. We do not train models on your data. The LLM your AI client uses is governed by your contract with that provider.
Where does my data go when I use MCP?▾
From Great Question to your AI client to the LLM provider you chose. We do not proxy MCP responses through any LLM provider ourselves.
What about PII?▾
A workspace-wide Hide PII via MCP toggle is on by default. Names, emails, and phone numbers are redacted at our server before responses leave. Admins can turn it off for approved workflows. The toggle is audit-logged.
Can I limit which users in my workspace can use MCP?▾
Today the toggle is workspace-wide. Per-user enable and disable is on the Q3 2026 roadmap. In the meantime, use RBAC to scope what each user can see. That scoping applies to MCP automatically.
Does it work with my AI client?▾
If it speaks MCP over HTTPS, yes. We officially test Claude Desktop, Claude Code, Cursor, ChatGPT, VS Code, Codex, Jules, Windsurf, Zed, Cline, Gemini, and Cowork.
Does it support audio or video?▾
Not via MCP. Text transcripts only.
Is there a separate DPA or sub-processor list for MCP?▾
No. Your existing Great Question DPA covers MCP. There are no MCP-specific sub-processors.
Can I run a trial before paying?▾
MCP is included on Pro and above. A trial or sandbox path is in exploration. Talk to your CS rep for early access.
Connect research to your AI tools
.png)
Five ways to engage
In order of commitment. Start with the one that fits where you are.
30-minute setup call
Customer Success walks you through your first three prompts on your actual repo. Free on Pro.
Book a call →Security review call
Engineering and Trust walk your CISO or IT team through architecture. SOC 2 reports under NDA.
Book a review →MCP-curious webinar
Monthly, with breakout rooms by role. 90 minutes. Open to ICs and leaders.
Register →MCP round table
Quarterly. Leaders only. Off the record. Heads of Research, Product, Design, CIOs. By invite.
Request invite →Want to be featured?
Doing something interesting with the Great Question MCP? We would love to put you on stage, write a customer story, or co-host a salon at your office.
Submit here →Your repo is already the best dataset you will never use. MCP is how you finally use it.
No metering. Five-minute install. Your existing permissions, your existing data, now in the AI tool your team already lives in.