> Documentation index: https://usertold.ai/llms.txt

# UserTold.ai

> Ask your users why, while they can still tell you.

## What it does

UserTold runs GPT-Live-1 voice Interviews with real participants inside your product — or silently observes usage (voice, screen, and navigation) — and returns source-linked Evidence: the quote or observed behavior behind each card, plus available screen and page context. GPT-Live-1 handles the natural full-duplex conversation while a backend agent follows the Study, uses grounded context, and calls approved tools. Related Evidence can support a Finding for review. Product triage decides whether a reviewed Finding becomes delivery work in Linear or GitHub. When a linked Linear issue completes, its Evidence resolves and later Evidence can prompt recurrence review.


## The research loop

1. **Study:** Install the project script once. Choose the research goal, interview flow, and pages where users can join.
2. **Interview:** With consent, record conversation and product use. Screen recording is available on supported desktop browsers.
3. **Evidence:** Review cards linked to what participants said or did, with recording timestamps and available page context.
4. **Finding:** Check the draft summary and its linked evidence, then mark the Finding reviewed.
5. **Delivery:** Choose which Findings to send to Linear for product triage or to GitHub for delivery.
6. **Follow-up:** Linear completion resolves linked evidence. Review similar evidence from later interviews for possible recurrence.

UserTold works with users your team can already reach. It does not recruit participants or guarantee responses.

## Start with users

Connect through [MCP](/docs/mcp), the [CLI](/docs/cli), or the [Quickstart](/docs/quickstart). Your agent can prepare an observational, task-focused, or open conversation study, then inspect the source-linked evidence returned after existing users choose to take part.

## MCP Server

Connect via Model Context Protocol for direct tool calls from Claude, Cursor, or any MCP-compatible agent:

```json
{
  "usertold": { "type": "http", "url": "https://mcp.usertold.ai/mcp" }
}
```

Auth: MCP clients open an OAuth 2.1 + PKCE browser login. See [MCP Integration](/docs/mcp) for setup details.

## Documentation

- [Quickstart](/docs/quickstart) — create a project, launch a study, and inspect the result
- [Core Concepts](/docs/concepts) — Projects, Studies, Intake, Evidence, Findings
- [Studies](/docs/studies) — interview script design
- [API Reference](/docs/api) — REST API
- [MCP Integration](/docs/mcp) — connect an MCP-compatible agent
- [CLI Reference](/docs/cli) — focused end-to-end workflow and live command discovery
- [Study Design Guide](/guides/study-design) — how to write effective study scripts
- [For Agents](/for-agents) — agentic loop integration

## Agent documentation

Fetch [/llms.txt](/llms.txt) for the complete index, or [/llms-full.txt](/llms-full.txt) for the setup and operations docs in one file.

## Research questions

Explore the [AI user interviewer](/features/ai-user-interviewer): observe a real task, prepare a running evidence summary when enabled, and conduct a planned voice debrief about what just happened.

Choose the product question you need to answer:

- [Understand onboarding drop-off](/guides/find-where-users-get-stuck).
- [Understand pricing confusion](/use-cases/pricing-page-research).
- [Discover the need behind a feature request](/guides/identify-user-needs).

Then [analyze the interviews](/guides/analyze-user-interviews) and [prioritize supported problems](/guides/prioritize-fixes-with-evidence).