Skip to main content

DevCycle MCP Server Documentation

Overview​

The DevCycle MCP (Model Context Protocol) Server enables AI coding assistants like Cursor and Claude to interact directly with DevCycle's feature flag management system. This integration allows developers to manage feature flags, variables, and targeting rules without leaving their coding environment.

DevCycle provides two MCP implementations:

  • Local MCP Server (dvc-mcp): Runs locally using CLI authentication or API keys
  • Remote MCP Server (https://mcp.devcycle.com): Hosted Cloudflare Worker with OAuth authentication

Both implementations share the same underlying tools, schemas, and utilities, ensuring consistent functionality regardless of which option you choose.

Table of Contents​

Installation & Setup​

Local MCP Server​

Prerequisites​

  • Node.js 18+ installed
  • DevCycle CLI installed globally: npm install -g @devcycle/cli
  • DevCycle account with API credentials or SSO authentication

Local Configuration​

For Cursor (.cursor/mcp_settings.json):

{
  "mcpServers": {
    "devcycle": {
      "command": "dvc-mcp"
    }
  }
}

For Claude Desktop:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "devcycle": {
      "command": "dvc-mcp"
    }
  }
}

Remote MCP Server (Cloudflare Worker)​

The remote MCP server provides OAuth-based authentication and requires no local installation.

Remote Configuration​

For Cursor (.cursor/mcp_settings.json):

{
  "mcpServers": {
    "devcycle": {
      "url": "https://mcp.devcycle.com/mcp"
    }
  }
}

For Claude Desktop:

{
  "mcpServers": {
    "devcycle": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.devcycle.com/mcp"
      ]
    }
  }
}

For more details about the remote MCP implementation, see the MCP Worker README.

Authentication​

Local MCP Server Authentication​

The local MCP server supports two authentication methods:

export DEVCYCLE_CLIENT_ID="your-client-id"
export DEVCYCLE_CLIENT_SECRET="your-client-secret"
export DEVCYCLE_PROJECT_KEY="your-project-key"
# Authenticate via SSO
dvc login sso

# Select your project
dvc projects select

Remote MCP Server Authentication​

The remote MCP server uses OAuth 2.0 authentication:

  • Users authenticate through DevCycle's Auth0 tenant
  • No API keys or local credentials needed
  • Project selection is handled through the select_project tool

Available Tools​

All tools listed below are available in both the local and remote MCP implementations. They share the same schemas, validation, and behavior.

Warning Symbols​

  • ⚠️ - This tool can affect production environments. The AI will confirm with you before executing.
  • ⚠️⚠️ - This tool performs destructive operations (deletions). The AI will require explicit confirmation before executing.

Feature Management​

list_features​

List all features in the current project with optional search and pagination.

Parameters:

  • search (optional): Search query to filter features (minimum 3 characters)
  • page (optional): Page number (default: 1)
  • perPage (optional): Items per page (default: 100, max: 1000)
  • sortBy (optional): Sort field (createdAt, updatedAt, name, key, createdBy, propertyKey)
  • sortOrder (optional): Sort order (asc, desc)
  • staleness (optional): Filter by staleness (all, unused, released, unmodified, notStale)
  • createdBy (optional): Filter by creator user ID
  • type (optional): Feature type (release, experiment, permission, ops)
  • status (optional): Feature status (active, complete, archived)

create_feature ⚠️​

Create a new feature flag. The AI assistant will guide you through a complete feature setup:

  1. Create a variable and associate it with the feature (defaults to a boolean variable with the same key)
  2. Create variations for the feature (defaults to "on" and "off" variations)
  3. Set and enable initial targeting for at least the development environment (defaults to all users with "on" variation)

Parameters:

  • key: Unique feature key (pattern: ^[a-z0-9-_.]+$)
  • name: Human-readable name (max 100 chars)
  • description (optional): Feature description (max 1000 chars)
  • type (optional): Feature type (release, experiment, permission, ops)
  • tags (optional): Array of tags for organization
  • variations (optional): Array of variations with key, name, and variables
  • configurations (optional): Environment-specific configurations
  • sdkVisibility (optional): SDK visibility settings
  • variables (optional): Array of variables to create or reassociate with this feature
  • controlVariation (optional): The key of the variation that is used as the control for Metrics
  • settings (optional): Feature-level settings configuration
  • interactive (optional): MCP-only hint to prompt for missing fields

update_feature ⚠️​

Update an existing feature flag.

Parameters:

  • key: Feature key to update
  • name (optional): New name
  • description (optional): New description
  • type (optional): New type
  • tags (optional): New tags
  • variations (optional): Updated variations
  • variables (optional): Updated array of variables for this feature
  • settings (optional): Updated feature-level settings configuration
  • sdkVisibility (optional): Updated SDK visibility settings
  • controlVariation (optional): Updated control variation key for Metrics

update_feature_status ⚠️​

Update the status of a feature flag.

Parameters:

  • key: Feature key
  • status: New status (active, complete, archived)
  • staticVariation (optional): Variation to serve if status is complete

delete_feature ⚠️⚠️​

Delete a feature flag from ALL environments.

Parameters:

  • key: Feature key to delete

cleanup_feature​

Fetch the DevCycle Feature Cleanup prompt and return its markdown content to guide safe cleanup of a completed feature and its variables in codebases.

Parameters:

  • featureKey: The feature key you plan to clean up (used for context in the prompt)

get_feature_audit_log_history​

Get feature flag audit log history from DevCycle. Returns audit log entities matching the DevCycle API schema with date, a0_user, and changes fields.

Parameters:

  • feature_key: Feature key
  • page (optional): Page number for pagination (default: 1)
  • perPage (optional): Number of items per page (default: 100, max: 1000)
  • sortBy (optional): Field to sort by (createdAt, updatedAt, action, user) (default: createdAt)
  • sortOrder (optional): Sort order (asc, desc) (default: desc)
  • startDate (optional): Start date for filtering (ISO 8601 format)
  • endDate (optional): End date for filtering (ISO 8601 format)
  • environment (optional): Environment key to filter by
  • user (optional): User ID to filter by
  • action (optional): Action type to filter by

Variable Management​

list_variables​

List all variables in the current project.

Parameters:

  • search (optional): Search query
  • page (optional): Page number
  • perPage (optional): Items per page
  • sortBy (optional): Sort field (createdAt, updatedAt, name, key, createdBy, propertyKey)
  • sortOrder (optional): Sort order (asc, desc)
  • feature (optional): Filter by feature key
  • type (optional): Variable type (String, Boolean, Number, JSON)
  • status (optional): Variable status (active, archived)

create_variable ⚠️​

Create a new variable.

Parameters:

  • key: Unique variable key (pattern: ^[a-z0-9-_.]+$)
  • type: Variable type (String, Boolean, Number, JSON)
  • name (optional): Variable name
  • description (optional): Variable description
  • defaultValue (optional): Default value
  • _feature (optional): Associated feature key
  • validationSchema (optional): Validation rules

update_variable ⚠️​

Update an existing variable.

Parameters:

  • key: Variable key to update
  • name (optional): New name
  • description (optional): New description
  • type (optional): New type
  • validationSchema (optional): New validation rules

delete_variable ⚠️⚠️​

Delete a variable from ALL environments.

Parameters:

  • key: Variable key to delete

Project Management​

get_current_project​

Get details of the currently selected project.

Returns the project defined by your current MCP context. If no project is selected:

  • Remote Worker: use the select_project tool
  • Local MCP: configure a project via dvc projects select or set DEVCYCLE_PROJECT_KEY

Parameters: None

select_project​

Select a project to use for subsequent MCP operations.

  • Local MCP: updates your local DevCycle configuration at ~/.config/devcycle/user.yml
  • Remote MCP: stores the selection in the worker state for your session

Parameters:

  • projectKey (optional): The project key to select. If not provided, lists available projects

SDK Installation​

install_devcycle_sdk​

Fetch DevCycle SDK installation instructions for your platform/language and follow the steps to install. Returns the corresponding guide markdown.

Parameters:

  • guide: Install guide identifier

Supported guide identifiers:

  • android, android-openfeature, angular, dotnet, dotnet-openfeature, flutter, go, go-openfeature, ios, ios-openfeature, java, java-openfeature, javascript, javascript-openfeature, nestjs, nestjs-openfeature, nextjs, nodejs, nodejs-openfeature, php, php-openfeature, python, python-openfeature, react, react-native, react-openfeature, roku, ruby, ruby-openfeature

Self-Targeting & Overrides​

get_self_targeting_identity​

Get current DevCycle identity for self-targeting.

Parameters: None

Returns your current DevCycle identity for the selected project, along with a dashboard link to profile overrides.

update_self_targeting_identity​

Update DevCycle identity for testing.

Parameters:

  • dvc_user_id: DevCycle User ID (use empty string to clear)

Updates your identity for self-targeting and testing. Use null or empty string to clear.

list_self_targeting_overrides​

List all active overrides for the current project.

Parameters: None

Lists all active overrides for the current project for your identity.

set_self_targeting_override ⚠️​

Set an override to test a specific variation.

Parameters:

  • feature_key: Feature key
  • environment_key: Environment key
  • variation_key: Variation to serve

Important: Confirm with the user before setting overrides in production environments.

clear_feature_self_targeting_overrides ⚠️​

Clear overrides for a specific feature/environment.

Parameters:

  • feature_key: Feature key
  • environment_key: Environment key

Important: Confirm with the user before clearing overrides in production environments.

Results & Analytics​

get_feature_total_evaluations​

Get total variable evaluations per time period for a specific feature.

Parameters:

  • featureKey: Feature key
  • startDate (optional): Start date as Unix timestamp (milliseconds since epoch)
  • endDate (optional): End date as Unix timestamp (milliseconds since epoch)
  • platform (optional): Platform filter for evaluation results
  • variable (optional): Variable key filter for evaluation results
  • environment (optional): Environment key to filter results
  • period (optional): Time aggregation period (day, hour, month)
  • sdkType (optional): Filter by SDK type (client, server, mobile, api)

get_project_total_evaluations​

Get total variable evaluations per time period for the entire project.

Parameters:

  • startDate (optional): Start date as Unix timestamp (milliseconds since epoch)
  • endDate (optional): End date as Unix timestamp (milliseconds since epoch)
  • platform (optional): Platform filter for evaluation results
  • variable (optional): Variable key filter for evaluation results
  • environment (optional): Environment key to filter results
  • period (optional): Time aggregation period (day, hour, month)
  • sdkType (optional): Filter by SDK type (client, server, mobile, api)

Error Handling​

The MCP server returns structured error responses:

{
  "errorType": "AUTHENTICATION_ERROR",
  "errorMessage": "401 Unauthorized",
  "toolName": "list_features",
  "suggestions": [
    "Re-authenticate with DevCycle (run \"dvc login sso\" for CLI for local MCP or re-login through OAuth for remote MCP)",
    "Verify your API credentials are correct",
    "Check if your token has expired"
  ],
  "timestamp": "2025-07-01T00:00:00.000Z"
}

Fields:

  • errorType: One of AUTHENTICATION_ERROR, PERMISSION_ERROR, RESOURCE_NOT_FOUND, VALIDATION_ERROR, SCHEMA_VALIDATION_ERROR, RATE_LIMIT_ERROR, NETWORK_ERROR, PROJECT_ERROR, UNKNOWN_ERROR.
  • errorMessage: Human-readable error description.
  • toolName: The MCP tool that produced the error.
  • suggestions: Remediation steps tailored to the error type.
  • timestamp: ISO 8601 timestamp when the error was generated.

Common error scenarios:

  • Authentication failures: Check credentials and project configuration
  • API rate limits: Implement retry logic in your automation
  • Validation errors: Ensure parameters meet requirements (patterns, lengths, etc.)
  • Permission errors: Verify your API key has necessary permissions

Usage Examples​

Creating a Feature Flag​

Create a feature flag for the new checkout flow with variations for A/B testing

The AI assistant will use:

  1. create_feature to create the feature, providing variables, variations, and initial environment configurations in one step

Viewing Analytics​

Show me the evaluation metrics for the checkout_flow feature over the last week

The AI assistant will use:

  1. get_feature_total_evaluations with appropriate date range
  2. Display the evaluation data grouped by time period

Managing Overrides for Testing​

Set up my identity to test the premium user experience

The AI assistant will use:

  1. update_self_targeting_identity to set your user ID
  2. set_self_targeting_override to force specific variations

Analyzing Feature Usage​

Show me the recent changes to the checkout_flow feature

The AI assistant will use:

  1. get_feature_audit_log_history to retrieve change history

Best Practices​

1. Production Safety​

Tools marked with ⚠️ can affect production environments. The AI will confirm before:

  • Creating or updating features/variables
  • Enabling/disabling targeting
  • Setting overrides in production

Tools marked with ⚠️⚠️ are destructive and require extra confirmation.

2. Naming Conventions​

  • Feature keys: lowercase_with_underscores or kebab-case
  • Variable keys: Follow the same pattern as features
  • Must match pattern: ^[a-z0-9-_.]+$
  • Maximum 100 characters

3. Environment Management​

  • Use consistent environment keys across projects
  • Common pattern: development, staging, production
  • Create environment-specific overrides for testing

4. Self-Targeting Best Practices​

  • Clear overrides after testing
  • Use meaningful user IDs for team collaboration
  • Document override purposes in team communications

5. Audit and Compliance​

  • Use get_feature_audit_log_history for compliance tracking
  • Tag features appropriately for organization
  • Include descriptions for documentation

Architecture & Implementation​

Both the local and remote MCP servers share:

  • Common Tools: All tool implementations in src/mcp/tools/
  • Schemas & Validation: Shared Zod schemas for input/output validation
  • API Clients: Both implement the IDevCycleApiClient interface
  • Error Handling: Consistent error messages and handling across implementations

Key differences:

  • Authentication: Local uses API keys/CLI auth, Remote uses OAuth 2.0
  • Transport: Local uses stdio, Remote uses SSE/HTTP
  • State Storage: Local uses file system, Remote uses Durable Objects

Notes on Environments​

  • Dedicated environment tools are not exposed via MCP. Use get_current_project or select_project to retrieve environments and SDK keys for the selected project.

Development & Local Testing​

For local development and testing of the MCP server:

Running from Source​

# Clone the repository
git clone https://github.com/DevCycleHQ/cli.git
cd cli

# Install dependencies
yarn install

# Build the project
yarn build

# Create a local package
npm pack

# Install the package globally (replace with actual filename)
npm install -g devcycle-cli-*.tgz

# Run the MCP server
dvc-mcp

Alternative: Direct Node Execution​

If you prefer to run directly without installing globally:

# After building, run the MCP server directly
node dist/mcp/index.js

Testing with AI Assistants​

For local testing, update your AI assistant configuration to point to the local build:

{
  "mcpServers": {
    "devcycle": {
      "command": "node",
      "args": ["/path/to/cli/dist/mcp/index.js"]
    }
  }
}

Debug Logging​

The MCP server logs all operations to stderr, which can be viewed in:

  • Cursor: Developer Tools console
  • Claude Desktop: Log files in the application support directory