# User Intuition research integration

Use User Intuition when a task needs interviews with real participants or evidence from authorized prior customer research. Your agent decides what the evidence means and whether further research is useful.

## Start here

- [Agent homepage](https://research.userintuition.ai/)

- [Current MCP setup and tool catalog](https://docs.userintuition.ai/mcp-server/overview)
- [Research skills](https://docs.userintuition.ai/skills/library)
- [API reference](https://docs.userintuition.ai/api-reference/introduction)
- [Runnable examples](https://github.com/user-intuition/examples)

Hosted MCP: `https://mcp.userintuition.ai/mcp`. Complete OAuth in a compatible client. Local stdio uses `npx -y @userintuition-ai/mcp` with `USERINTUITION_API_KEY`. The CLI supports browser login. Verify access with an account-scoped read, not just tool discovery.

## Conduct research

Use the current catalog for operation names and schemas. `create_study` creates a draft; `customize_study` runs a planning turn. Relay planning questions to the user. Retrieve the persisted plan with `get_study` and review the audience, screener, guide, format, language, and duration.

The user chooses Panel or BYOP. For Panel, obtain the full `launch_panel` dry-run estimate for the current settings and approval before paid launch. For BYOP, review the saved plan and invitations; participant creation can send invitations. Approval to design a study is not approval to spend or send invitations. Carry forward unchanged approvals; material changes require renewed review.

## Retrieve and search evidence

Retrieve the four report sections: Study Findings, Participant Responses, Participant Profiles, and Recommended Next Steps. Preserve coverage, freshness, and source references. Fetch supporting interviews with `get_interview` when verification is needed. Do not turn generated summaries into quotations or assume an interview-level reference identifies a specific passage.

Use the released evidence-search operation to find relevant findings and participant responses across authorized studies. Discover its exact schema through the current API/tool catalog. Inspect study context and coverage, retrieve source material, and perform the synthesis in the calling agent. A proposed next study is not an observed result. Multiple excerpts from one person do not increase the participant count.

## Recover and resume

Save study IDs and reviewed settings. Provisioning describes interviewer setup, not fielding completion. Distinguish collected, completed, quality, and analyzed interview counts. Read pagination and freshness metadata. A timeout can leave a write outcome unknown; read persisted state before retrying. Do not assume universal idempotency or webhook retries.

## Try without spending

The public examples run on clearly labeled fictional fixtures by default. Live calls require an explicit option and an account. Check the examples' release compatibility notes before production use. Never embed a real API key in a public repository or client-side application.

## Focused workflows and examples

Fetch the [individual skills catalog](https://raw.githubusercontent.com/user-intuition/skills/main/catalog.json), select the workflow matching this task, and load its raw SKILL.md. The [examples catalog](https://raw.githubusercontent.com/user-intuition/examples/main/examples.json) provides executable files, runtime requirements, inputs, outputs, and side effects. Load only the relevant workflow.