LaunchDarkly MCP Server

The LaunchDarkly MCP Server is LaunchDarkly's official Model Context Protocol implementation, connecting MCP-capable AI clients such as Claude Code, Cursor, VS Code with GitHub Copilot, or Windsurf directly to the LaunchDarkly API. Instead of managing feature flags, targeting rules, or rollout percentages exclusively through the web dashboard, an agent can be instructed in plain language to create a flag, adjust a targeting rule, or check the status of a gradual rollout. The repository lives under the official launchdarkly GitHub organization, is licensed under MIT, and was last updated on September 24, 2026 — an actively maintained project run by LaunchDarkly itself rather than a community wrapper.

Hosted server recommended, local server for EU and Federal

LaunchDarkly offers the server in two official variants. The hosted server at https://mcp.launchdarkly.com/mcp/launchdarkly is, according to the provider, the recommended path: sign-in happens through OAuth consent rather than a hardcoded API key, and it is officially documented for Cursor, Claude Code, VS Code with GitHub Copilot, and Windsurf. For customers on an EU or Federal instance of LaunchDarkly, the hosted server is not available per the documentation — the provider explicitly points these customers to the self-managed local server from the same repository. That local server can run via npx, as a downloaded standalone binary for several platforms, from a cloned source checkout, or as a Docker image through the AWS Marketplace or an ECR registry. This entry treats both paths as a single server (deployment_type = both) because they come from the same official repository and expose the same tools.

Tools for feature flags, AI Configs, and observability

The server groups its tools into several resource categories. The Feature Flags area covers full lifecycle management: creating, reading, patching, and deleting flags, plus checking targeting rules and status per environment. AI Configs let an agent create and manage configurations for AI models and variations directly. The server additionally exposes the Audit Log for traceability, Code References to link flags to repositories, and Environments to list environments per project. According to the provider, observability tools are also included, letting an agent query logs, traces, error groups, and dashboard data through chat, alongside preconfigured "Agent Skills" for multi-step procedures such as flag cleanup with an automatic safety check before deletion.

Installation and authentication

For the hosted server, visiting the install page at https://mcp.launchdarkly.com/mcp/launchdarkly/install is often enough to connect without a manual configuration file; alternatively, the endpoint can be added manually to a client's MCP configuration. For the local server, the documented command is npx -y --package @launchdarkly/mcp-server -- mcp start --api-key <TOKEN>, where the API key is generated through the authorization page in the LaunchDarkly account and authenticates against https://app.launchdarkly.com (standard), https://app.launchdarkly.us (Federal), or https://app.eu.launchdarkly.com (EU) depending on the instance. For write access to feature flags, LaunchDarkly's own documentation recommends at minimum the Writer base role or the predefined Developer preset role, not an account with full administrator rights.

Cost and prerequisites

The MCP server's source code itself is free under the MIT license. The underlying LaunchDarkly platform offers, per its own pricing page, a permanently free Developer tier with unlimited seats and core features, alongside usage-based and individually negotiated Enterprise tiers for larger teams. A LaunchDarkly account with an appropriately scoped API key is required either way; current quotas and plan details should be checked on the official pricing page, since they can change.

Who benefits from the LaunchDarkly MCP Server

The server suits development teams that already manage feature flags, staged rollouts, or A/B tests through LaunchDarkly and want to handle recurring work — creating a new flag, adjusting a targeting rule, checking rollout progress, or cleaning up stale flags — directly from a coding agent instead of switching between editor and dashboard. It is especially useful for teams building AI-assisted deployment and release workflows who want flag changes tied closely to the surrounding code context. Anyone not using LaunchDarkly, or managing flags solely through their own internal tooling, gets no benefit from this server.

Security and limits

Because feature flags directly control the behavior of production applications, an agent using this server should get only the minimum necessary permissions — LaunchDarkly explicitly recommends a restricted role instead of full administrator rights. Write actions such as deleting flags or changing targeting rules in production environments should first be tested on non-critical flags or in a test environment before being applied more broadly. When running locally, treat the API key like any other access secret and avoid sharing it in plain text or committing it to a repository; the hosted server's OAuth consent flow handles this safeguard for you.

An MCP-compatible AI client (e.g. Claude Code, Cursor, VS Code with GitHub Copilot, or Windsurf), a LaunchDarkly account with access to the target project, and for local use an API key with the Writer or Developer role.

Hosted (recommended): open the install page https://mcp.launchdarkly.com/mcp/launchdarkly/install and complete the OAuth consent flow, or add the endpoint manually to the client configuration. Local (e.g. for EU/Federal): run npx -y --package @launchdarkly/mcp-server -- mcp start --api-key <TOKEN> with an API key generated through the LaunchDarkly authorization page.

npx -y --package @launchdarkly/mcp-server -- mcp start --api-key <TOKEN>

Hosted server: OAuth consent on connect. Local server: a personal LaunchDarkly API key passed via the --api-key parameter, scoped to the relevant instance (standard, Federal, or EU).

LaunchDarkly recommends at minimum the Writer base role or the Developer preset role for write access, rather than full administrator rights.

The server returns flag, targeting, audit log, and observability data to the connected AI client, which may forward results to its model provider; review that provider’s data policy separately.

Feature flags control the behavior of production applications. Test write changes on non-critical flags first, grant only the minimum necessary permissions, and treat the local API key like any other secret.