@growthbook/mcp

Exposes GrowthBook's feature flag evaluation engine through the Model Context Protocol, allowing Claude and other MCP-compatible clients to query feature flag states, variations, and targeting rules without direct API calls. Implements MCP resource and tool handlers that translate client requests into GrowthBook SDK method calls, maintaining session context across multiple evaluations within a single conversation.

Provides MCP tools to fetch experiment configurations, targeting rules, and variation assignments from GrowthBook without requiring direct REST API calls. Implements resource handlers that serialize GrowthBook experiment objects into structured JSON, exposing rule conditions, audience targeting, and variation weights for inspection and decision-making within AI workflows.

Enables MCP clients to evaluate feature flags with arbitrary user attributes and context, implementing a schema-based parameter handler that maps user context objects to GrowthBook SDK evaluation calls. Supports custom attributes, user IDs, and environment-specific context, allowing Claude to simulate flag behavior for different user segments without hardcoding evaluation logic.

Registers GrowthBook feature flags and experiments as MCP resources, making them discoverable and queryable by Claude through the MCP resource protocol. Implements resource URI schemes (e.g., growthbook://flag/{id}, growthbook://experiment/{id}) that map to GrowthBook SDK objects, allowing Claude to reference and inspect flags/experiments as first-class entities within conversations.

Handles GrowthBook SDK initialization and credential management at MCP server startup, accepting configuration through environment variables or constructor parameters. Implements a single-instance SDK pattern where credentials are loaded once and reused across all MCP tool/resource calls, eliminating per-request authentication overhead and maintaining consistent evaluation state.

Provides MCP tools to query which variation a user is assigned to in an A/B test and retrieve test results/metrics from GrowthBook. Implements evaluation logic that returns both the assigned variation and associated metadata (rule matched, timestamp, experiment ID), enabling Claude to understand test outcomes and make data-driven decisions based on experiment results.