Base44
Build and edit Base44 apps from an AI assistant, and query data your app's users submit.
GrowthBook MCP
Read and change GrowthBook feature flags and experiments from an AI agent.
Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.
Paste it into any MCP client. Setup docs
What the GrowthBook MCP connector does
The GrowthBook MCP server connects GrowthBook feature flagging and experimentation to Claude and any MCP-compatible agent. It exposes four tools: two that load GrowthBook's bundled Agent Skills, and two that make authenticated read and write calls against the REST API. The competence lives in the skills.
Connection verified
Connection checked by Agentman on .
Listed in Anthropic’s Connectors Directory at community tier — a listing, not a partner-level review.
Anthropic states this reflects the level of review a connector received, not a security audit.
GrowthBook MCP tools (4)
- growthbook_api_read
- growthbook_api_write
- growthbook_list_skills
- growthbook_read_skill
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- The API tools do not validate payloads. GrowthBook states this directly; the guardrails are in the skills, and skipping them means calling your REST API unchecked.
growthbook_api_writecovers DELETE, so the write surface is the whole REST API, not a short list of named mutations.- No documented read-only mode. Capability-only mode removes the skills, not the write tool.
offline_accessis in the grant, so authorisation persists beyond one session.- 1.x tools are gone. Per-endpoint tools were removed in 2.x; configurations expecting them break.
- We did not read tool schemas. The endpoint requires OAuth, so names come from Anthropic's directory and behaviour from GrowthBook's documentation. We exercised no tool.
Frequently asked questions
Why does the GrowthBook MCP server have only four tools?
Because it is a thin bridge, not an endpoint wrapper. GrowthBook describes the server as loading Agent Skills and making authenticated calls to its REST API. Two tools list and read those skills; the other two are authenticated GET and mutating passthroughs to the API. Version 2.x deliberately removed the old per-endpoint tools.
Does the GrowthBook MCP server validate what an agent sends?
No, and GrowthBook says so plainly: the API tools are authenticated passthroughs that do not validate payloads. It places the guardrails in the skills instead, which encode workflows like draft then review then publish and safe rollouts. An agent that skips the skills is calling your REST API unguarded.
Can I stop the GrowthBook MCP server changing anything?
Not through a documented read-only mode. growthbook_api_write performs POST, PUT, PATCH and DELETE against the REST API, and there is no separate read-only endpoint. GrowthBook does document a capability-only mode at slash mcp slash api that drops the skills, but that exposes both read and write tools, not read alone.
What is the GrowthBook capability-only mode for?
Teams that supply their own workflows. Connecting to slash mcp slash api instead of slash mcp exposes only growthbook_api_read and growthbook_api_write, without the bundled skills. GrowthBook suggests it when your client or team already has its own guardrails, so you are not loading GrowthBook's opinions about how to sequence changes.
Do I need an API key for GrowthBook Cloud MCP?
No. GrowthBook documents that Cloud users sign in with OAuth in the browser with no API key in the MCP config, and we confirmed the server issues an OAuth challenge naming api.growthbook.io as its authorization server. The advertised scopes are openid, profile, email and offline_access.
I used GrowthBook MCP 1.x and my tools disappeared. Why?
Version 2.x removed them. GrowthBook documents that the old per-endpoint tools such as create_feature_flag and get_experiments are gone, replaced by the skill loader plus the API read and write tools. Cloud clients should point at mcp.growthbook.io slash mcp, and local installs should upgrade to the latest package.
Sources
- GrowthBook MCP server documentation (retrieved 2026-09-13), the URL the server itself advertises as
resource_documentation.docs.growthbook.io/robots.txtcarriesContent-Signal: ai-train=yes, search=yes, ai-input=yes— an explicit publisher opt-in to AI synthesis, recorded here because we rely on it. · retrieved 2026-09-13 - Live OAuth posture check against
https://mcp.growthbook.io/mcp— HTTP 401 with awww-authenticateBearer challenge carryingrealm="GrowthBook MCP", and RFC 9728 metadata athttps://mcp.growthbook.io/.well-known/oauth-protected-resource/mcpreturning fourscopes_supportedandhttps://api.growthbook.ioas the authorization server (2026-09-13) · retrieved 2026-09-13 - Anthropic Connectors Directory entry (retrieved 2026-09-13) · retrieved 2026-09-13
- GrowthBook support — <support@growthbook.io> · Privacy
More Developer Tools connectors
CircleCI
Diagnose failed builds, read job logs, inspect test results and rerun or roll back deployments from your AI assistant.
Claude Managed Agents
Create, run, monitor and debug Anthropic's Claude Managed Agents from inside Claude itself.
Product names and logos belong to their owners. Agentman is not affiliated with a connector's maker unless the page says the connector is built and run by Agentman.
Use it in an agent
Put GrowthBook MCP to work.
Connect GrowthBook MCP once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.