All directories
SkillGrowthProduct

GrowthBook Skills + MCP

Official GrowthBook agent skills and MCP server — create flags, run experiments, conclude winners, and clean stale flags from Claude Code, Cursor, or VS Code.

Open on GitHubBrowse more skills
01

Overview

GrowthBook’s agent-native surface lets coding agents operate the full feature-flag and experimentation lifecycle without leaving the editor. Install `growthbook/skills` for REST-backed playbooks (setup, flag revisions, review, publish, cleanup, experiments) or connect `@growthbook/mcp` for tool calling. Same permissions, approvals, and audit logs as the GrowthBook app. Pairs with A/B Testing Frameworks and Analytics skills for hypothesis design plus warehouse-native measurement.

SaaSE-commerceMarketplaceOther

Work with LimeDock

Using this skill? LimeDock can wire it into a durable automation you own.

Skills show what's possible. LimeDock builds and runs the owned marketing, sales, and ops automations around them.

We sell owned automations for SaaS teams — live workflows that plug into Slack, CRM, and your internal platform — not just a skill list.

Book a workflow call
02

Link

https://github.com/growthbook/skills
03

Installation guide

# Agent skills (REST helper, no MCP required)
npx skills add growthbook/skills

# MCP server (Cursor / VS Code / Claude Code)
npx -y @growthbook/mcp@latest

Claude Code MCP:

claude mcp add growthbook --transport stdio \
  --env GB_API_KEY=YOUR_API_KEY \
  --env GB_EMAIL=YOUR_EMAIL \
  -- npx -y @growthbook/mcp@latest

Requires a GrowthBook API key (Settings → API Keys). Optional: `GB_API_URL` / `GB_APP_ORIGIN` for self-hosted.

04

How to use it

1. Run `gb-setup` (or set env vars) so the agent can reach your instance. 2. Ask to create a flag, wrap a UI surface, or draft an experiment from org defaults. 3. Use revision skills for draft → review → publish. 4. Query experiment summaries; ship winners; run stale-flag cleanup. 5. Keep humans in the approval path for production environments.

05

Example prompts

  • “Create a boolean feature flag `new-checkout-flow` that defaults to false and wrap the pricing CTA.”
  • “Create an A/B test on `new-checkout-flow` with control false and treatment true using our defaults.”
  • “Summarize the results of my last 5 GrowthBook experiments.”
  • “Show me which feature flags are stale right now and propose cleanup.”
06

Use cases and examples

  • Create a boolean flag and wrap a checkout CTA from the editor
  • Spin up an A/B test linked to a feature flag with metric defaults
  • Summarize last N experiment results without leaving Cursor
  • Find and archive stale flags after a cleanup sprint
  • Ramp a search-ranking flag from 1% → 100% over two weeks
07

Prerequisites

  • GrowthBook Cloud or self-hosted instance + API key
  • Claude Code, Cursor, VS Code Copilot, or Codex with MCP/skills support
  • Node.js for `npx` MCP install
  • Permission model: use a PAT scoped to what the agent should touch
08

Tips

  • Skills call the REST API via a bundled helper; MCP is optional but nicer for interactive tool use.
  • Self-hosted: set GB_API_URL / GB_APP_ORIGIN; use GB_HTTP_HEADER_* behind Cloudflare Access.
  • Pair with `ab-testing` for hypothesis/sample-size design before creating experiments.
  • Agent actions inherit the same RBAC and approval workflows as humans.
09

Sources