The customer-research API for AI agents
Your agent designs the study, recruits real people, runs the interviews, and returns analyzed insight — 72 MCP tools, no dashboard required.
Trusted by teams at
AI agents running customer research autonomously — commissioning interviews, recruiting respondents, and returning analyzed insight — are becoming standard infrastructure for product and insights teams. User Intuition is the MCP server that gives any AI agent access to this infrastructure: 72 tools across 9 capability groups, from study design to analyzed results. User Intuition's AI moderator runs voice, chat, and video interviews with real panelists, probing 5-7 layers deep so agents receive decision drivers, not just surface preferences. Dashboard-only tools require a human to log in and configure each step; an MCP-native research interface removes that bottleneck for agent workflows entirely. A User Intuition agentic study starts at $200, returns machine-readable results in 24-48 hours via a ui_sk_ API key, and draws from a 4M+ global panel in 50+ languages. User Intuition returns preference splits, agreement scores, ranked themes, and minority objections with verbatim quotes — every interview compounding in the Intelligence Hub for the next agent query.
Why Agents Can't Use Dashboard-Only Research Tools
Most research platforms were built for humans opening a browser. Four structural gaps make them incompatible with agent-native workflows.
Your agent can't open the dashboard
Qualitative research tools are designed for human clicks: log in, configure a study, wait for results, export a PDF. An AI agent can't do any of that. Without a programmatic interface, the agent has to hand off to a human at every step — eliminating the autonomy that makes agentic workflows valuable.
Survey APIs are not moderated interviews
Survey-distribution APIs return static responses to static questions. They can't probe why a respondent chose Option A, surface the minority objection that explains churn, or ladder down from a stated preference to the decision driver beneath it. AI-moderated interviews produce the depth that survey logic can't.
Scraped data is not real human signal
Web scraping, social listening, and review mining tell you what people said publicly, not how they react to your specific content today. An agent building a message test or concept validation needs fresh, targeted human reactions — not a corpus of historical mentions.
LLM inference collapses variance
Asking an LLM to simulate audience reactions flattens the real distribution: the 15% who reject your claim and the 52% who love it get averaged into one confident answer. Real participants surfacing genuine skepticism, confusion, and emotional responses are the only source of evidence an agent can trust for high-stakes decisions.
What the MCP Server Gives Your Agent Instead
What matters most to teams after switching to AI-moderated research.
Every step of the research workflow — study creation, participant recruitment, interview analysis, report generation — exposed as MCP tools an agent can call directly
AI-moderated voice, chat, and video interviews with vetted panelists, returning genuine preference splits, agreement rates, and minority objections no LLM can fabricate
From ask_humans call to structured results while the decision window is still open — no export, no PDF parsing, no human relay required
Every study automatically feeds the Intelligence Hub so agents can query accumulated research history, not just the latest run
What Is the User Intuition MCP Server?
The User Intuition MCP server is the full User Intuition research platform exposed as 72 MCP tools — callable directly from Claude, ChatGPT, Cursor, Claude Code, VS Code, or any agent that supports the Model Context Protocol. One API key. No dashboard login. Your agent designs the study, recruits participants, runs AI-moderated interviews, and retrieves structured results programmatically.
How Does an AI Agent Run Customer Research?
An AI agent runs customer research by calling User Intuition's MCP tools directly: create a study via ask_humans or create_assistant, recruit participants from a 4M+ global panel or your own list, wait for AI-moderated interviews to complete, then retrieve preference splits, agreement scores, themes, and verbatim quotes via get_results or get_call. No dashboard login, no manual export — the agent owns the full loop.
Which AI clients are supported?
Claude Desktop, Claude Code, Cursor, and VS Code use the stdio transport — install with npx -y userintuition-mcp and set USERINTUITION_API_KEY. ChatGPT uses the Streamable HTTP transport with OAuth. Any agent framework that supports the open Model Context Protocol (MCP) standard can connect.
What is the difference between Human Signal and Studies?
Human Signal (5 tools) is the fast, paid-panel path: your agent calls ask_humans with a mode (preference, claim, or message), specifies stimuli and sample size, and gets back a structured result in hours. Studies (13 tools) is the full interview-workflow path: the agent creates a custom study via create_assistant, configures screeners and moderation prompts, manages participant invites, and triggers transcript analysis. Both return machine-readable results; Human Signal is optimized for speed and directional signal, Studies for depth and custom design.
Do I need a User Intuition account?
Yes. Sign up at app.userintuition.ai, then generate an API key from Settings > API Keys. The key takes the form ui_sk_... and is the only credential the MCP server requires for stdio mode. ChatGPT uses OAuth instead of a direct key.
The Full User Intuition Research Platform via MCP
Every capability is a tool an agent can call. No dashboard required for any of them.
Human Signal (5 tools)
Create and manage paid panel studies that ask real people what they think — preference checks, claim reactions, and message tests returning structured results in hours.
Studies (13 tools)
Build, configure, and manage full AI-moderated interview studies end-to-end — create assistants, set screeners, upload concept links, and control panel surveys.
Invites & Participants (8 tools)
Recruit from your own customer list or the RepData panel — create individual or bulk invites, manage participant records, and send Tremendous rewards on completion.
Calls & Interviews (7 tools)
Access transcripts, recordings, and analysis for every completed interview — list calls, fetch individual transcripts, update visibility, and trigger study-level report generation.
Voice & Reports (2 tools)
Select from the catalog of available interviewer voices and retrieve the latest AI-generated analysis report for any study.
Intelligence Hub (18 tools)
Search and synthesize all accumulated research — query the file-search store, manage sessions and chat history, and generate reports or PowerPoints from the full evidence base.
Integrations & Panels (5 tools)
Sync customer segments from Shopify or HubSpot, list external participants, check integration status, and provision RepData panel surveys for a study.
Monetization & Utilities (8 tools)
Manage your wallet, browse subscription and credit plans, redeem coupon codes, and handle referral invitations — all programmatically from your agent.
Account (6 tools)
Retrieve organization details and member lists, update your profile, submit feedback on a study, and contact sales or support — without leaving your agent workflow.
Connect Your AI Agent in Minutes
One-time MCP setup. Works with any compatible client — no dashboard login required after setup.
Get a ui_sk_ API Key
Sign up at app.userintuition.ai and generate an API key from Settings > API Keys. The key takes the form ui_sk_... and is the only credential the MCP server requires for stdio mode.
Connect Your Client
Claude Desktop / Cursor / Claude Code / VS Code: add {"userintuition": {"command": "npx", "args": ["-y", "userintuition-mcp"], "env": {"USERINTUITION_API_KEY": "ui_sk_..."}}} to your MCP config. ChatGPT: use the Streamable HTTP transport with OAuth — run the server with --transport streamable-http and point ChatGPT to your server URL.
Your Agent Calls a Tool
The agent picks the right tool for the job: ask_humans for a quick Human Signal study, or create_assistant for a full AI-moderated interview workflow. Specify mode, stimuli, sample size, and audience — the server handles the rest.
Real Interviews Run
Participants join AI-moderated voice, chat, or video conversations. The AI moderator probes 5-7 layers deep to separate stated preferences from decision drivers. Use the dry_run flag first to preview cost and timeline before committing.
Analyzed Results Return
Call get_results or get_assistant_report to retrieve preference splits, agreement scores, ranked themes, minority objections with verbatim quotes, and a data quality score. Every study automatically feeds the Intelligence Hub for future queries.
Agent-Native Research vs. Dashboard-Only Tools
vs. Data & Scraping APIs
| Dimension | User Intuition MCP | Dashboard-Only Tools | Data & Scraping APIs |
|---|---|---|---|
| Agent access | Native MCP — 72 tools, no dashboard login | Human must log in and configure each study | API exists but returns historical or scraped data |
| Interview depth | AI-moderated conversations with 5-7 layer laddering | AI-moderated or human-moderated, but dashboard-gated | No interviews — static data or social content |
| Real people | Yes — 4M+ vetted panel or your own list | Yes — but manual export blocks agent consumption | Depends — some panels exist, no moderation |
| Result format | Structured JSON — agent-ready, no parsing | PDF or dashboard UI — not agent-consumable | Raw text, ratings, or embeddings — requires post-processing |
| Fresh signal | On-demand — agent triggers a new study any time | On-demand — but requires human setup | Historical or batch — not specific to your content today |
| Cost | From $200 per study, $20/interview | Varies — often $5K–$15K+ per project | Varies — often cheap per record, but low validity |
| Compounding memory | Every study feeds Intelligence Hub | Standalone reports, not queryable by agents | No organizational memory layer |
What Agents Use the MCP Server For
See how agent-driven workflows map to specific research solutions.
Concept & Message Testing
Agent calls ask_humans in message or preference mode to validate copy, positioning, and creative before launch.
→Win-Loss Analysis
Agent creates a win-loss study via create_assistant using the Win/Loss study type to understand the real reasons deals are won or lost.
→Brand Health Tracking
Agent schedules recurring Brand Health studies to track perception and competitive positioning over time without manual setup.
→NPS + CSAT Deep Dives
Agent triggers NPS+CSAT study type interviews with detractors to surface the drivers behind scores — not just the numbers.
→Consumer Insights
Agent runs preference checks and claim-reaction studies to uncover purchase motivations and unmet needs at scale.
→User Research
Agent commissions usability and concept-test studies with your own customer list via bulk_create_invites_from_segment.
→When Should an Agent Use Human Signal vs. a Full Study?
Human Signal (ask_humans) is optimized for speed and directional signal — preference checks, claim reactions, and message tests returning results in hours. Full studies (create_assistant) are better for deep exploration, complex audience segmentation, and board-level deliverables.
Use Human Signal When
- You need quick signal on messaging or creative before launch
- Comparing headlines, taglines, or product name options
- Checking whether a claim feels believable to your audience
- Testing if messaging is clear and lands the way you intend
- Running iterative test-and-revise cycles inside an agent workflow
- You need directional validation in hours, not days
Use Full Studies When
- Deep exploratory research requiring 30+ minute AI-moderated conversations
- Custom screeners and audience segmentation beyond panel defaults
- Concept testing with external links or video stimuli
- Board-level deliverables with full evidence trails and PowerPoint output
- Longitudinal tracking using the same study design over weeks or months
- Recruiting from your own customer list via Shopify or HubSpot segments
Both Human Signal and full studies feed the same Intelligence Hub — findings compound regardless of which tools created them.
"We were about to launch a rebrand with copy our AI helped write. Ran a message test first — 24% of respondents found the tagline confusing. We caught a $200K mistake in 3 hours for less than the cost of lunch."
VP of Marketing — Series B SaaS, 150 employees
Frequently Asked Questions
Related resources
Pillar Guides
Deep-dive guides covering this topic from strategy to execution.
Tools & Tactics
Practical frameworks and platform-specific guides for teams ready to act.
Reference Guides
Reference deep-dives on methodology, best practices, and applied research.
Alternatives & Comparisons
Side-by-side comparisons with competing platforms and approaches.
Related Solutions
Complementary research use cases that pair with this topic.
Industries
See how teams in specific verticals apply this research.
Add Real Human Signal to Every AI Decision
Get an API key and connect your agent in under 5 minutes, or explore the docs first.
Works with Claude, ChatGPT, Cursor, Claude Code, VS Code, and any MCP-compatible agent.
Last updated