With the GrowthBook MCP server, you can interact with GrowthBook right from your LLM client. Manage feature flags, run experiments, analyze results, and more.
Setup
Environment Variables
Variable
Status
Description
GB_API_KEY
Required
A GrowthBook API key or PAT. When using a PAT, MCP server capabilities are limited by its permissions.
GB_EMAIL
Required
Your email address used with GrowthBook. Used when creating feature flags and experiments.
GB_API_URL
Optional
Your GrowthBook API URL. Defaults to https://api.growthbook.io.
GB_APP_ORIGIN
Optional
Your GrowthBook app URL. Defaults to https://app.growthbook.io.
GB_HTTP_HEADER_*
Optional
Custom HTTP headers for all API requests. Pattern: GB_HTTP_HEADER_<NAME>=value where underscores become hyphens. Example: GB_HTTP_HEADER_X_TENANT_ID=abc123 sends X-Tenant-ID: abc123.
Add the MCP server to your AI tool of choice. See the official docs for a complete guide.
Tools
Feature Flags
Tool
Description
create_feature_flag
Create a new feature flag (disabled in all environments by default)
get_feature_flags
List flags or fetch details for a specific flag by ID
list_feature_keys
Get all feature flag IDs (keys only), optionally filtered by project
get_stale_feature_flags
Check flag staleness and get cleanup guidance
generate_flag_types
Generate TypeScript types for feature flags
update_feature_flag
Update flag properties (description, defaultValue, tags, etc.)
toggle_feature_flag
Enable or disable a flag per environment
create_force_rule
Add a targeting rule to a flag (all environments)
add_feature_rule
Add a targeting rule to a specific environment
reorder_feature_rules
Set rule evaluation order for an environment
remove_feature_rule
Remove a rule from an environment
Experiments
Tool
Description
create_experiment
Create a new A/B test (call get_defaults first)
get_experiments
List experiments or fetch details (metadata, summary, or full modes)
Change targeting on a running experiment without flipping status (defaults to a new phase to keep analysis clean)
resume_experiment
Resume a stopped experiment back to running with optional new targeting (appends a fresh phase)
stop_experiment
Stop a running experiment, optionally declaring a winner
archive_experiment
Archive or unarchive an experiment
refresh_experiment_results
Trigger a fresh analysis snapshot and return results (supports dimension and phase)
get_attributes
List user attributes for targeting
Metrics
Tool
Description
get_metrics
List all metrics (fact and legacy)
create_fact_metric
Create a proportion, mean, quantile, or ratio fact metric
update_fact_metric
Update an existing fact metric
delete_fact_metric
Delete a fact metric
get_data_sources
List data sources or fetch one with assignment query SQL
list_fact_tables
List available fact tables (SQL data sources)
create_fact_table
Create a new fact table
update_fact_table
Update an existing fact table
delete_fact_table
Delete a fact table
list_fact_metrics
List fact metrics with configuration details
list_fact_table_filters
List filters for a fact table
create_fact_table_filter
Create a filter on a fact table
delete_fact_table_filter
Delete a fact table filter
Dimensions
Tool
Description
list_dimensions
List dimensions available for experiment breakdowns
create_dimension
Create a new dimension for result segmentation
delete_dimension
Delete a dimension
Configuration & Search
Tool
Description
get_environments
List all environments (production, staging, etc.)
get_projects
List all projects in your organization
get_sdk_connections
List SDK connections
create_sdk_connection
Create a new SDK connection
get_defaults
Get default config for creating experiments
set_user_defaults
Set custom experiment defaults
clear_user_defaults
Reset to automatic defaults
search_growthbook_docs
Search official GrowthBook documentation
Development
Prerequisites
Node.js (LTS)
npm
Setup
git clone https://github.com/growthbook/growthbook-mcp.git
cd growthbook-mcp
npm install
npm run build
Commands
npm run build # TypeScript compilation → server/
npm run dev # Watch mode
npm test # Run all tests
npm run test:coverage # Tests with coverage
npm run generate-api-types # Regenerate types from OpenAPI spec
Architecture
src/
index.ts # Entry point, registers all tool groups
utils.ts # Shared utilities (fetch, headers, rate limiting)
format-responses.ts # Response formatters (agent-friendly markdown)
api-types.d.ts # Auto-generated from OpenAPI spec (DO NOT EDIT)
api-type-helpers.ts # Typed aliases for API responses
tools/
features.ts # Feature flag tools
experiments/
experiments.ts # Experiment lifecycle tools
metrics.ts # Metrics and fact table tools
dimensions.ts # Dimension tools
defaults.ts # Experiment defaults
environments.ts # Environment listing
projects.ts # Project listing
sdk-connections.ts # SDK connections
search.ts # Documentation search
prompts/ # MCP prompt registrations
Each tool group exports a register*Tools() function called from index.ts. All tools use Zod schemas for input validation and return agent-friendly markdown via formatters in format-responses.ts.
Contributing
See CLAUDE.md for detailed development guidelines including:
