Open-source warehouse-native experimentation.

What you can do via MCP

Example prompts the agent runs.

• “Set up an A/B test for the new pricing page with 50/50 traffic split.”

  Creates the experiment in GrowthBook with the right targeting and primary metric, returns the SDK snippet for the integration.

• “Wrap every new feature in a flag, gradually deploy.”

  Generates a per-feature flag config plus a rollout schedule the agent follows over days or weeks.

• “Identify and clean up stale flags. Schedule weekly.”

  Pulls the flag list, scores by last evaluation, opens a PR removing dead flag references, and registers the cleanup as a recurring task.

• “Give me an update on my banner test for the pricing page.”

  Fetches results, summarizes lift and confidence, calls out segment differences, and recommends a next step.

• “Analyze all past checkout flow experiments. Find top 3 win/lose reasons.”

  Reads experiment history, clusters outcomes, and ships a brief on what consistently moves checkout completion.

# GrowthBook OSS MCP, local server
# Run: GROWTHBOOK_API_KEY=key_abc123 npx growthbook-mcp

mcp.growthbook.create_experiment({
  trackingKey: "pricing-page-v2",
  name: "Pricing page A/B",
  hypothesis: "New tier hierarchy increases paid conversion.",
  variations: [
    { id: "0", name: "Control", weight: 0.5 },
    { id: "1", name: "Treatment", weight: 0.5 }
  ],
  metrics: ["paid_signup"],
  datasource: "snowflake_prod"
})

# Runs SQL on your warehouse; results stay in your stack

Visual demonstration

What this looks like in practice.

Pairs with these skills

The PM-experimentation skill suite.

This integration pairs with the forthcoming experiment-design, feature-flagging skills. The skill landing pages and SKILL.md sources land in subsequent dispatches; cross-link hyperlinks are added when the skill pages ship.