GrowthBook MCP Server

AI agent · Coding agent · Engineering & development by GrowthBook

Bring your AI agents into your release process with MCP-native feature flags and experimentation.

GrowthBook MCP Server is a bridge connecting AI coding agents to GrowthBook's experimentation platform. It allows developers to manage feature flags and analyze experiments directly from IDEs like Cursor and Claude Code while maintaining human-in-the-loop governance.

Agent specs

The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.

Languages
English

Category: AI Agents

Channels
API
Integrations
GrowthBook Cloud\nGrowthBook Self-Hosted\nREST API
Editors and tools
Claude Code\nCursor\nCodex\nVS Code\nClaude Desktop
Deployment
Cloud (hosted URL), Docker, Local process (stdio)
Protocol
Model Context Protocol (MCP)
Supported Agents
Claude Code, Cursor, VS Code, Codex, Claude Desktop
  • Connects AI agents to GrowthBook via Model Context Protocol
  • Enforces human-in-the-loop approval workflows for all agent actions
  • Enables agents to query analytics and conclude experiment winners
  • Automates cleanup of stale feature flags through AI workflows

Overview

GrowthBook MCP Server is an open-source bridge that connects AI coding agents to the GrowthBook experimentation and feature flagging platform. By exposing the platform's REST API to agentic workflows, it enables developers to manage releases and testing directly from their code editors. The server ensures that all agent actions are governed by the same role-based permissions and approval flows as human team members. It is designed to facilitate AI-native development by bringing experimentation data into the AI's context. The server can be deployed via GrowthBook Cloud with OAuth or as a local process for self-hosted instances.

Experimentation and Flag Tasks

  • Agentic Flag Creation: AI agents can create new feature flags and experiments directly within the coding environment without switching to the GrowthBook application.
  • Rollout Configuration: Agents can configure ramp schedules for feature flags, such as increasing traffic from 1% to 100% over a specified period.
  • Analytics Querying: The server enables agents to query product metrics and analyze how key indicators like Daily Active Users (DAU) change after a launch.
  • Automated Code Cleanup: AI assistants can identify and archive stale feature flags to maintain code cleanliness and reduce technical debt.

Governance and Control

  • Human-in-the-Loop Governance: Every action taken by an AI agent is subject to the same approval workflows and role-based permissions as human team members.
  • Review Queue Integration: Agent drafts are sent to the team's review queue for verification before being published to production.
  • Audit Logging: All agent activities are recorded in audit logs to maintain a clear history of changes and ensure accountability.
  • Environment Scoping: Agents are restricted by environment scoping to ensure they only operate within authorized development or production stages.

Integration and Deployment

  • Cloud Connectivity: Users can connect to GrowthBook Cloud by pointing their client at a specific URL and signing in via OAuth.
  • Self-Hosted Deployment: For self-hosted GrowthBook instances, the server supports connection via stdio or an official Docker image.
  • API Bridge Architecture: The server acts as a thin bridge that loads bundled Agent Skills and calls the GrowthBook REST API on the agent's behalf.

What it does

  • Create feature flags\nRun experiments\nConfigure rollouts\nQuery analytics\nConclude winners\nClean up stale code

Use cases

  • Developers use AI agents to create and configure feature flags for new search ranking algorithms without leaving their IDE.
  • Engineering teams set up automated ramp schedules from 1% to 100% traffic using AI prompts.
  • Data analysts query experiment results through an AI assistant to see how DAU changed after a new feature launch.
  • Product teams use agents to identify and archive stale feature flags to reduce technical debt automatically.
  • Teams enforce security by requiring all AI-generated flag drafts to pass through human approval workflows.

Specifications

Channels
API
Integrations
GrowthBook Cloud\nGrowthBook Self-Hosted\nREST API
Editors and tools
Claude Code\nCursor\nCodex\nVS Code\nClaude Desktop
Human control
Agent drafts land in team review queue; actions governed by approval workflows and RBAC
Deployment
Cloud (hosted URL), Docker, Local process (stdio)
Setup
Connect in minutes by adding MCP server URL or running npx command
Protocol
Model Context Protocol (MCP)
Supported Agents
Claude Code, Cursor, VS Code, Codex, Claude Desktop
Deployment Options
Cloud, Docker, Local process (stdio)
Authentication
OAuth (Cloud), API Key (Self-hosted)
Governance Features
RBAC, Audit logs, Approval workflows, Environment scoping
Primary Interface
REST API Bridge
Installation Command
npx @growthbook/wizard

How it works

  1. Generate API Key: Sign up for a GrowthBook account or connect to a self-hosted instance to generate a required API key from account settings.
  2. Connect the Agent: Add the GrowthBook MCP server URL to your agentic client or install the skills via npx to enable guardrailed workflows.
  3. Prompt and Execute: Use natural language prompts within your editor to ask the AI assistant to create flags, check experiments, or analyze data.
  4. Review and Approve: Review the agent's drafts in the GrowthBook review queue and approve them to roll out changes to production.

FAQ

What is the difference between the MCP server and the Agent Skills plugin?

The MCP server is a thin bridge that loads skills and calls the REST API, ideal for MCP-native hosts like Claude Desktop. Agent Skills are markdown playbooks installed locally for agents that can run shell commands like Claude Code.

Which integration method should I choose?

If your agent cannot run shell commands or a terminal (like Claude Desktop), you should use the MCP server. If you use Claude Code and prefer slash commands, the Agent Skills plugin is recommended.

Can I use the MCP server with a self-hosted GrowthBook instance?

Yes, the MCP server supports self-hosted instances through stdio or by using the official GrowthBook Docker image.

How do I authenticate the MCP server on GrowthBook Cloud?

GrowthBook Cloud users can sign in securely using OAuth after pointing their MCP client to the hosted server URL.

Start for Free

Added by AntSpace from public sources (growthbook.io).

Related

More AI agents by GrowthBook