LaunchDarkly's MCP server is the official implementation of the Model Context Protocol for LaunchDarkly. It exposes an MCP interface that enables AI clients to interact with LaunchDarkly data such as AI configurations (aiConfigs), feature flags, environments, and code references. The server defines a set of resources and operations including aiConfigs with targeting management (getTargeting, updateTargeting, list, create, delete, get, update) and variation management (createVariation, deleteVariation, getVariation, updateVariation); codeReferences with listRepositories; environments with listByProject; and featureFlags with status and CRUD-like operations (getStatus, list, create, get, patch, delete). The project provides integration instructions for multiple clients (Cursor, Claude, Qodo Gen) and supports standalone binaries, running from a local clone, or via Docker. API access is controlled with an API key obtained from LaunchDarkly's Authorization page, and the server can be started using mcp start commands (for example via npx @launchdarkly/mcp-server -- mcp start --api-key ...). The README also covers environment-variable-based tokens, and notes that some AI clients may not support env vars directly. The server can connect to different LaunchDarkly environments (Commercial, Federal, EU) via the --server-url setting.