Agentic Research MCP Integration Quickstart

Agentic research works through the Model Context Protocol (MCP) — an open standard that lets AI platforms connect to external tools. Once connected, your AI agent can launch consumer studies, receive structured results, and query accumulated intelligence without leaving your workflow. The integration brings consumer evidence into the same surface where the work is already happening, instead of forcing a context switch into a separate research tool.

This quickstart covers the setup for the three most common platforms — Claude, ChatGPT, and Cursor — and explains what your agent can actually do once the connection is live. For broader context on the methodology behind the studies your agent will launch, see AI Customer Interviews: The Complete Guide. User Intuition studies start at $200, return results in 24-48 hours, and run across a 4M+ panel covering 50+ languages.

What is MCP and why does it matter for research?

The Model Context Protocol is an open standard that lets AI assistants discover and call external tools in a consistent way. Before MCP, every integration was bespoke: a different API client for every assistant, a different authentication flow, a different way of declaring what tools the assistant could call. MCP collapses that into a single protocol, which is why the same User Intuition server works from Claude, ChatGPT, and Cursor with only the connection-side configuration changing.

For research, the practical effect is that the AI assistant becomes a first-class research client. The assistant can launch a preference check during a strategy conversation, retrieve results when they land, and query the Customer Intelligence Hub for context on prior findings — all without the researcher opening a separate tool. The boundary between “AI conversation” and “consumer evidence” disappears. A product manager debating a feature in Claude can ask 20 customers in the same conversation. A copywriter testing a headline in ChatGPT can pull 15 real reactions. A developer pricing a new tier in Cursor can validate the format with SMB buyers without leaving the IDE.

This is what makes agentic research different from traditional API integration. The AI assistant is not just a thin wrapper over the research platform; it is a research collaborator that decides when to ask the customer and how to interpret the answer in the context of the conversation it is already having with the user.

Prerequisites

Before starting, you need:

1. A User Intuition account. Sign up here (Starter tier is free, three free interviews on signup with no card required).
2. API key. Generated from your User Intuition dashboard under Settings > API Keys.
3. An MCP-compatible AI platform. ChatGPT, Claude, or Cursor, plus any other MCP-compatible client (Continue, Zed, etc.) using the same configuration pattern.
4. A network path that allows outbound HTTPS to the User Intuition MCP server endpoint shown in your dashboard. Enterprise environments behind a corporate proxy may need to allow-list the endpoint before the first connection succeeds.

Setup: Claude (MCP native)

Claude has native MCP support, making it the most straightforward integration.

Step 1: Open Claude and navigate to Settings > MCP Servers.

Step 2: Add a new MCP server with the User Intuition server URL from your dashboard.

Step 3: Authenticate with your API key. Paste the key into the credential prompt; Claude stores it securely in the platform’s credential store, not in plain text.

Step 4: Test the connection by asking Claude: “What consumer research tools do I have access to?”

Claude should respond by listing the available research modes (preference check, claim reaction, message test) and the Customer Intelligence Hub query capability. If the response is a generic apology, the connection did not land — re-check the server URL and API key, and confirm your account has at least one credit available.

Setup: ChatGPT (app integration)

User Intuition is available as a ChatGPT App.

Step 1: In ChatGPT, search for “User Intuition” in the app/plugin store.

Step 2: Install and authenticate with your User Intuition API key.

Step 3: Test by asking ChatGPT: “Run a preference check with 10 people comparing these two headlines: ‘Ship faster’ vs. ‘Build smarter.’”

The ChatGPT integration uses the same underlying MCP server as Claude. The difference is the credential management UI and the way ChatGPT surfaces the tool list to the model. Functionally, the same study types, the same panel, and the same pricing apply.

Setup: Cursor (IDE integration)

Cursor’s MCP support lets developers run consumer research from their IDE — useful when the question is about a piece of copy, a feature flag, or a pricing variant the developer is actively building.

Step 1: Open Cursor Settings > MCP.

Step 2: Add the User Intuition MCP server configuration. The MCP block goes in your mcp.json or the equivalent settings UI.

Step 3: Authenticate with your API key.

Step 4: Test from your code context: “Ask 10 SMB buyers whether they prefer monthly or annual pricing display on this page.”

Cursor’s value-add is that the agent has the code context loaded already. A question about a pricing display can include the actual rendered copy from the file the developer is editing; a question about a feature flow can include the screen states the agent can see. The integration shortens the loop between “I wrote this thing” and “real users reacted to this thing” from days to hours.

What is the cost model for running studies through the MCP integration?

The cost model is identical to the platform’s web UI: $20 per audio interview, $10 per chat interview, $40 per video interview, with studies starting at $200 and a 4M+ panel across 50+ languages. The MCP integration does not add a surcharge; the integration is the way many teams consume the product, not a premium tier on top of it.

This matters for budgeting because the agentic pattern unlocks higher study frequency than teams running the platform only through the dashboard. When an agent can launch a 15-person preference check during a product strategy conversation in Claude, the team ends up running 5-10x more small studies per year than they would have if every study required a researcher to open the dashboard separately. The per-study cost is unchanged; the volume of studies the team finds it worth running goes up substantially.

The Professional tier ($999/month) includes 50 free credits per month, which absorbs most small-study volume for a team using the integration heavily. At 50 free audio interviews per month, a team can run roughly 3-5 small studies (10-15 participants each) without burning into pay-per-use credits. Larger studies and high-volume teams burn into credit balance from there, with extra credits at the same $20/$10/$40 rates the Starter plan uses.

How does the MCP integration handle authentication and scope?

Authentication runs on a per-workspace API key generated from the User Intuition dashboard. The key carries the workspace’s permissions and credit balance; multiple keys can exist per workspace for team members who want isolated authentication, and a key can be revoked at any time without affecting the rest of the workspace.

Scope-wise, the MCP server exposes four tool categories to the connected agent: study launch, study status and retrieval, intelligence hub query, and account/credit management. The agent does not get raw database access; it gets a tool surface that mirrors what a human researcher would do through the dashboard. This is what lets the same key safely connect from Claude, ChatGPT, and Cursor simultaneously — the tools, not the data, are what the agent can reach.

Enterprise workspaces can scope keys further: read-only keys that can query the hub but not launch new studies, study-launch-only keys that cannot read prior research, or keys restricted to a single project. The scoping is set in the workspace settings and applies to every MCP connection that uses the key. This matches the typical enterprise pattern of giving a development team a read-only key for prototyping while reserving study-launch authority for the research team.

Comparing the three integrations

All three connect to the same panel, the same study types, and the same intelligence hub. The choice is about which surface the team is already working in, not about which gives access to different research capabilities.

What can your agent actually do once connected?

Once connected, launch your first study with a natural language prompt:

Preference check: “Compare these two taglines with 15 marketing directors at CPG brands: ‘Insights that compound’ vs. ‘Research at the speed of decisions.’ Tell me which one resonates more and why.”

Claim reaction: “Test this claim with 20 SaaS product managers: ‘AI-moderated interviews achieve 98% participant satisfaction.’ Is it credible? What objections come up?”

Message test: “Test this homepage copy with 15 enterprise software buyers: ‘Customer intelligence that compounds. Every conversation builds on the last.’ Is it clear? What do people think it means?”

The agent handles study setup, participant recruitment, AI-moderated conversations, and analysis. Results arrive in 2-3 hours for small studies (10-15 participants), 24-48 hours for standard studies (20-50 participants), and up to 72 hours for larger studies (100+ participants). For modality detail, see AI interview modalities: voice vs video vs chat; the agent defaults to audio at $20/interview but the team can override to chat ($10) or video ($40) per study.

The agent also handles result interpretation. When the study completes, the agent does not just dump the transcripts into the chat — it summarizes the key themes, surfaces the strongest verbatim quotes, flags the divergence between segments, and links each conclusion back to the underlying interviews so the user can drill in. The interpretation runs against the same methodology the platform’s web UI uses, so the output is comparable regardless of which surface the user is in.

How do you query the intelligence hub from your agent?

After running studies, you can query accumulated findings:

• “What have we learned about pricing perception in the SMB segment?”
• “Summarize all competitive mentions from the last 3 months of studies.”
• “What messaging themes have tested best with enterprise buyers?”
• “Show me every verbatim where a participant mentioned a workaround they built because the product was missing a feature.”
• “What concerns came up across the last five concept tests in the financial services audience?”

The intelligence hub draws on all prior studies to answer these questions — making your agent more informed with every study you run. The compounding effect is the actual point of the integration. The first study answers a question. The tenth study answers the question and contextualizes it against nine prior studies. The fiftieth study answers the question, contextualizes it against 49 prior studies, and surfaces patterns that no individual study could have shown. The hub is what turns a single research project into a continuous intelligence asset, which is the opposite of the knowledge-decay pattern that slide-deck research follows.

What are the most common troubleshooting issues?

The most common issues fall into four buckets, in roughly the order they appear during setup.

Connection fails: Verify your API key is active in the User Intuition dashboard. Regenerate if needed. Check that the MCP server URL in your client matches the URL shown in the dashboard exactly — a trailing slash or stray protocol prefix is the most common cause.

Study doesn’t launch: Check that your account has sufficient credits. Starter tier requires pay-per-use credits; Professional tier includes 50 monthly. The error message returned to the agent will name the constraint explicitly (out of credits, panel not available for the requested segment, etc.).

Results delayed: Large studies (30+ participants) may take 24-72 hours. Small studies (10-15) typically complete in 2-3 hours. If a study is materially overdue against this window, the dashboard’s study detail page shows the participant fielding state and any recruitment friction.

Hub queries return nothing: The intelligence hub draws on completed studies. Run at least one study before querying. A new account will get an empty hub on the first query because there is nothing in it yet — this is expected behavior, not an error.

MCP server timeout: Most often a configuration path error in the client. Re-check the server URL, the credential scope, and any corporate proxy that might be intercepting the HTTPS connection. The agent’s verbose-mode output usually names the layer that failed.

When should you reach for the MCP integration vs. the web dashboard?

The two surfaces serve different research moments. The MCP integration is the right surface when the question arrives in the middle of an AI-assisted conversation — a product manager drafting a feature spec in Claude, a copywriter iterating on a tagline in ChatGPT, a developer building a checkout flow in Cursor. The cost of context-switching out to a separate dashboard kills most of these in-flight questions; the integration lets the question convert into a study without breaking the conversation.

The web dashboard is the right surface when the research moment is deliberate and the study is large enough that the setup deserves a dedicated workspace. Multi-segment studies, longitudinal trackers, complex screener logic, and large studies with custom recruitment all benefit from the dashboard’s full UI, which gives the researcher more control over every part of the configuration than the agent’s natural-language interface can express. The intelligence hub is also more navigable in the dashboard; the agent’s hub queries are excellent for targeted questions, but a researcher exploring patterns across hundreds of past studies will move faster in the dashboard’s UI.

Most teams end up using both. The agentic surface handles the in-flight questions; the dashboard handles the deliberate research program. The hub is the same in both places.

What does this all look like in one paragraph?

Agentic research via MCP turns any compatible AI assistant into a first-class research client that can launch consumer studies, retrieve results, and query the intelligence hub from inside the conversation the user is already having. Setup takes under 10 minutes on Claude (MCP native), ChatGPT (app integration), or Cursor (IDE MCP), and the same User Intuition API key works across all three platforms because the protocol is the same. Once connected, the agent can run preference checks, claim reactions, and message tests against a 4M+ panel covering 50+ languages, with small studies completing in 2-3 hours and standard studies in 24-48 hours at $20 per audio interview. The intelligence hub query capability is the compounding asset: every study the agent launches adds to a knowledge base that supports cross-study questions, so a team that has run 50 studies through the integration has a hub that can answer questions no single study could. The integration removes the context switch between AI-assisted work and consumer evidence, which is the binding constraint on most research-informed decisions today.

For full documentation, visit docs.userintuition.ai/integrations/mcp-server. For related reference: agentic research compliance and security covers the procurement and security review, agentic research vs. traditional qual decision matrix covers when to reach for an agentic study vs. a longer traditional engagement, and evidence trails for auditable customer intelligence covers the audit logging that enterprise reviewers expect. Studies start at $200, return results in 24-48 hours, and carry 5/5 ratings on G2 and Capterra.