Run user research from your coding agent

Pick the tier that matches how much signal you need:

Instant and Live are free and available now. Recruited-panel estimates work today; paid ordering is rolling out — see Credits & pricing.

1. Register the server. One line — nothing to clone or install. It runs straight from npm. No API key needed up front; you'll approve access in the browser.

Claude Code

claude mcp add userdelphi -s user \
  --env UD_BASE_URL=https://userdelphi.vercel.app \
  -- npx -y userdelphi-mcp

Cursor / Codex (mcp.json)

{
  "mcpServers": {
    "userdelphi": {
      "command": "npx",
      "args": ["-y", "userdelphi-mcp"],
      "env": { "UD_BASE_URL": "https://userdelphi.vercel.app" }
    }
  }
}

2. Restart your editor. MCP servers load when the editor starts — not when you add them — so quit and reopen (or start a new session). In Claude Code, run /mcp to confirm userdelphi is connected and showing its tools. The first launch downloads the package, so give it a few seconds.

3. Connect. Ask your agent to “connect to UserDelphi.” It prints a short code and a link — open the link, sign in, approve the code, and the key is saved locally. That's it.

Prefer to manage the key yourself? Create one under Settings → API keys and pass it as UD_API_KEY instead of connecting.

Once connected, your agent has these tools. You rarely call them by name — just describe what you want and the agent picks the right one.

Instant and Live studies are free. Synthetic personas cost nothing, and a share link to your own users costs nothing. You only pay to recruit real, matched participants.

Recruited panels are priced in credits (1 credit = 1 US cent). A study's price is built from a few factors:

• Base — about $8 for a 5-minute prototype session with no targeting.
• Session length — longer sessions cost more (5, 15, 20, 30 min).
• Targeting — each requirement you add (region, occupation, industry, age, language…) adds a surcharge, capped at 3×.
• Volume — 5%–20% off as you order more participants (5+, 10+, 20+, 50+).
• Study type — task-based studies cost a bit more than open prototype feedback. There's a $25 order minimum.

Estimates are always free. Ask your agent “what would 10 recruited testers cost for this?” and it returns an itemised quote. For example, 10 testers · 20-minute session · targeting occupation + industry comes to $362.88 ($36.29 per completed session). You only ever pay for sessions that are actually delivered — unfilled slots are refunded.

Instant feedback in a minute

You: Get feedback on my new checkout — https://staging.myapp.com/checkout
Agent: Running an instant study with 5 AI personas…
       Done. 3 of 5 missed the promo-code field — it reads as
       "apply discount". Here's the full report: …

Price a real panel before ordering

You: What would 10 real testers in tech, 20-min sessions, cost?
Agent: $362.88 for 10 completed sessions ($36.29 each) —
       targeting occupation + industry. Want me to order it?

Pull the report later

You: What's the top issue from the checkout study?
Agent: The promo-code field. 6 of 8 participants tried to use it
       as the coupon entry and got confused. Recommended fix: …