Growthbook MCP Server
Feature flag and skill management through GrowthBook API read/write tools.
Data last scanned today · Reviewed today
Overview
GrowthBook MCP Server provides direct access to GrowthBook API operations and bundled skill workflows. It allows AI clients to list and read specific skill entry points, or perform authenticated read/write operations against the GrowthBook API. The server includes tools for both standard API interactions and specific skill-related workflows, enabling programmatic management of features and configurations directly within an AI environment. The server supports two transport modes: standard stdio and HTTP. It handles authentication via API keys or personal access tokens and is designed to allow clients to distinguish between read-only and destructive actions through dedicated API tools.
Our verdict
This server carries a high risk level with a security score of 43/100. Because it is a community-maintained project rather than an official vendor release, and because critical security aspects such as the authentication method have not yet been editorially reviewed, users should exercise caution. Given its capability to perform write operations via `growthbook_api_write`, it is primarily suited for development environments where the implications of API-level changes are fully understood.
- Bundled skill definitions available for targeted workflow execution.
- Supports distinct API read and write tool segregation.
- Minimal dependency footprint with only three direct dependencies.
- Configurable for both stdio and HTTP transport layers.
- Community-maintained project lacking official vendor oversight.
- Authentication mechanisms remain awaiting expert editorial review.
- Security score of 43/100 signals significant unvetted risk surface.
Setup
{
"mcpServers": {
"growthbook-mcp": {
"args": [
"-y",
"@growthbook/mcp"
],
"command": "npx"
}
}
}Paste into ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows), then restart Claude Desktop. Some servers require API keys or tokens as environment variables — check the project's README and add an "env" object if needed.
- No read-only mode — running this grants full read/write access.
- No built-in authentication — anyone with access to this config can invoke every tool it exposes.
Tools
| Tool | Description | Risk |
|---|---|---|
| growthbook_list_skills | List top-level skill entry points (name + description). | low |
| growthbook_read_skill | Return a listed skill or qualified child workflow. | low |
| growthbook_api_read | Authenticated GET passthrough to the GrowthBook API. | low |
| growthbook_api_write | Authenticated POST/PUT/PATCH/DELETE passthrough. | high |
Compatibility
| Client | Local | Docker | Remote | Read-only |
|---|---|---|---|---|
| ChatGPT | ||||
| Claude Desktop | ||||
| Cursor | ||||
| VS Code | ||||
| Windsurf |
Frequently asked questions
›What configuration is required to run the server?
You must provide a 'GB_API_KEY' (GrowthBook API key or personal access token) and optionally specify the 'GB_API_URL' if using a self-hosted instance.
›Can I use this server without exposing skill tools?
Yes, you can disable skill tools by setting the environment variable 'GB_SKILLS_ENABLED' to 'false' or '0'.
›How do I restrict access to only API read and write tools?
You can point your MCP client to the '/mcp/api' endpoint if running in HTTP mode, or set 'GB_SKILLS_ENABLED=false' when using stdio.
›How are skills updated in this server?
Skills are bundled at build time; to update them, you must update the commit hash in 'agent-skills.lock.json' and rebuild the package.
›Does the server use the gb-call shell adapter?
No, the server does not shell out to 'gb-call'; instead, you map those commands to the provided 'growthbook_api_read' and 'growthbook_api_write' tools.
Changelog
Badge
Maintain this server? Add the live badge to your README.