WorldClone is a market validation engine that simulates user reactions pre- and post-launch with AI panels that mirror real-world distributions, helping teams de-risk and iterate faster.

How does the AI panel work?

We generate AI persona panels that reflect real population distributions, then run concept, price, and message tests instantly. Every result carries a confidence grade calibrated against public backtests.

PMs, growth marketers, brand teams, and research teams at B2B companies who need to validate their products before launch.

Published on npm · free install

PUBLIC DECISION MCP

Turn a fuzzy question into the next experiment

Use it from your local AI agent without a WorldCloneLab account or service key. Create a directional memo, label its trust ceiling, and record the real outcome.

No additional WorldCloneLab usage fee. Your Claude Code or Codex subscription and usage policies still apply.

No account or token

No WorldCloneLab login, access token, or database key.

Saved in your workspace

Memos and pilot drafts stay under .local/decisions.

Honest refusal

Unsupported decisions return unsupported and suggest real research.

Connect in one line

Available

claude mcp add --scope user worldclonelab -- npx -y @worldclonelab/decision-mcp

Three steps to a first decision

Connect the MCP

Run the command and confirm worldclonelab in your MCP list.

Choose a recipe

Ask: “Use worldclone_list_recipes and show me the copy recipe.”

Close with reality

Run the memo’s pilot and record the result with worldclone_record_outcome.

GUIDED RECIPES

Start even if you do not know what to ask

Pick one decision per session and fill the minimum intake with your agent.

Copy A/B

Choose which two variants deserve a real A/B first.

Price range

Create a synthetic band without treating it as a real survey.

First ICP

Compare two audiences for the same product and sequence the pilot.

Offer structure

Compare trials, diagnostics, guarantees, or packages.

Acquisition channel

Sequence fixed-budget channel pilots.

Value proposition

Narrow the first claim to test in market.

Share a decision

Prepare a reviewable card with trust and caveats attached.

The public package contains only these six tools

worldclone_list_recipes

Explains minimum inputs and trust ceilings.

worldclone_decide

Creates a Decision Memo and real-test draft.

worldclone_forecast_segments

Compares ICP or segment priorities.

worldclone_get_decision

Reopens saved memos and artifacts.

worldclone_record_outcome

Records the real result as self_reported.

worldclone_prepare_community_share

Prepares a share card without uploading it.

KOREAN PERSONA DATA

Korean persona data

Uses NVIDIA's public Korean persona dataset, Nemotron-Personas-Korea. One command downloads it to your machine, ready to use.

No account or token

Downloads without a HuggingFace account or API token.

Attribution included

CC BY 4.0 attribution is saved alongside as ATTRIBUTION.md.

Size guide

Embeddings are about 8 MB per 5,000 rows. The embedding model cache is larger.

Download command

npx -y @worldclonelab/decision-mcp@latest corpus pull --limit 5000

The @latest tag is pinned because a cached npx run can stick to an old version.
Add --skip-embeddings to download only the distribution data, without embeddings.
The corpus pull command ships from decision-mcp 0.3.0.

Local first, explicitly shared

The MCP sends no silent usage telemetry to WorldCloneLab. The share tool only prepares Markdown and a URL fragment. Data reaches the server only after you open, review, and post it.

Browse examples and known limits See backtest evidence