synthpanel

Who is this for?

Three jobs synthpanel does well — pick the one closest to yours.

See it in action

A run_quick_poll against three personas, with the auto-synthesis at the bottom. This is the raw shape of what you get back.

# Question
"Would you pay $29/month for a tool that runs synthetic focus groups?"

── Sarah Chen · 34 · Startup PM ───────────────────────────
Honestly? Yes, for a quarter. $29 is under my no-approval-needed
threshold, and if it saves me one botched launch that's already paid
back. I'd want to see at least one real-world validation case first.
verdict: yes  confidence: 0.7

── Marcus Patel · 41 · Senior UX Researcher ───────────────
Not as a replacement for recruited studies, but as a pre-filter — yes.
$29 is cheap enough that I'd expense it personally. My concern is bias:
I need to know how the personas were selected before I trust the
synthesis.
verdict: conditional  confidence: 0.6

── Priya Okafor · 29 · AI Engineer ────────────────────────
I'd pay it to skip building the scaffolding myself, but I'd want an
API, not just a CLI. If it plugs into my agent via MCP I'm in at $29,
probably $99 if the SDK is clean.
verdict: yes  confidence: 0.8

── Synthesis ──────────────────────────────────────────────
3/3 lean toward yes at $29, but each attaches a condition:
  • PM wants a validation case study before committing
  • Researcher wants transparency on persona selection / bias
  • Engineer wants SDK + MCP parity, signals $99 ceiling
Consensus: price is not the blocker; trust + integration depth are.
themes: price-fit, bias-transparency, sdk-parity, mcp-integration

Example output, formatted for readability. Real results are returned as structured JSON via CLI or MCP tool call.

MCP Server

Give your AI coding assistant access to synthetic focus groups. Drop this config into your editor and start running panels from chat.

// Claude Code · Cursor · Windsurf · Zed
{
  "mcpServers": {
    "synth_panel": {
      "command": "synthpanel",
      "args": ["mcp-serve"],
      "env": { "ANTHROPIC_API_KEY": "sk-..." }
    }
  }
}

Requires pip install synthpanel[mcp]. 12 tools: run prompts, run panels, manage persona & instrument packs, and more.

Quick start

# 1. install
pip install synthpanel

# 2. add a key — env var (one-shot) or stored (persistent)
export ANTHROPIC_API_KEY="sk-..."             # env var
# or: synthpanel login --provider anthropic --api-key sk-...   # persisted

# 3. one-shot prompt against the default model
synthpanel prompt "What do you think of the name Traitprint?"

# 4. run a full panel and save the result
synthpanel panel run \
  --personas examples/personas.yaml \
  --instrument examples/survey.yaml \
  --save

# 5. render a shareable Markdown report from the saved result
synthpanel report <result-id> -o report.md

More commands: synthpanel pack calibrate (calibrate a persona pack against a SynthBench baseline), synthpanel instruments (manage branching instrument packs), synthpanel analyze (statistics on saved results), synthpanel results (list and inspect saved runs by ID), synthpanel cost (spend summary), synthpanel login / whoami (credential management). Run synthpanel --help for the full list.

Every synthpanel report output opens with a mandatory synthetic-panel banner — “Synthetic panel. All responses below were generated by AI personas, not human respondents. Do not cite as user-research data.” — so the rendered Markdown can’t be mistaken for real-user research. Markdown v1 only; HTML deferred to v2. See the sp-viz-layer spec.

Measured against real humans

The SynthBench benchmark’s ground truth is real survey respondents — the General Social Survey (NORC), the Pew American Trends Panel (via the OpinionsQA, SubPOP, and GlobalOpinionQA datasets), and the World Values Survey — and every number below is recomputable from public data and open code. GSS is public-use, so for it the real human answer shares are shown directly.