LaunchDarkly MCP Server

Official LaunchDarkly MCP server for managing feature flags, AI configs, targeting rules, environments, and observability data from an AI client.

Developer Tools by LaunchDarkly OAuth2 (hosted) or API Key (local) active

Connect (Remote Server) GitHub Docs npm: @launchdarkly/mcp-server

Overview

LaunchDarkly's MCP integration gives AI clients direct access to LaunchDarkly's feature management, AI Configs, and observability APIs. It is offered in two flavors: three hosted MCP servers (Feature Management, AI Configs, Observability) authenticated via OAuth at mcp.launchdarkly.com, and a self-hosted open source server (@launchdarkly/mcp-server) that runs locally via npx using an API key.

The Feature Management server exposes tools for creating, listing, patching, and deleting flags, checking flag status, and inspecting environments, audit log entries, and code references. The AI Configs server provides full CRUD over AI Configs and variations plus targeting management. The Observability server adds queries against logs, traces, sessions, error groups, and dashboards. The hosted servers are the recommended path, with the local server useful for development, restricted networks, or custom permission scopes.

LaunchDarkly also publishes "agent skills" (launchdarkly/agent-skills) that wrap multi-step workflows on top of the MCP tools for Claude Code, Cursor, VS Code, Windsurf, and Codex.

Tools

Tool	Description
`Feature Flags: list / get / create / patch / delete / getStatus`	Manage feature flags in a project: list flags, fetch a single flag's configuration, create new flags, apply JSON patch updates (toggle, targeting, variations), delete flags, and check flag health/status.
`AI Configs: list / get / create / update / delete`	CRUD operations over LaunchDarkly AI Configs, which manage model selection, prompts, and parameters for AI features.
`AI Configs: createVariation / getVariation / updateVariation / deleteVariation`	Manage variations of an AI Config (e.g. different models or prompt templates).
`AI Configs: getTargeting / updateTargeting`	Read and update targeting rules that decide which AI Config variation is served to which users.
`Environments: listByProject`	List all environments configured for a given LaunchDarkly project.
`Audit Log: listEntries`	Fetch audit log entries for changes made across projects, flags, and configurations.
`Code References: listRepositories`	List repositories connected to LaunchDarkly for code reference scanning.
`Observability tools (hosted server)`	Query logs, traces, sessions, error groups, and create/inspect dashboards via the observability MCP endpoint.

Setup Guide

Option 1: Hosted MCP servers (recommended)

LaunchDarkly hosts three MCP endpoints authenticated via OAuth. Add any or all of them to your MCP client config:

{
  "mcpServers": {
    "launchdarkly-feature-management": {
      "type": "http",
      "url": "https://mcp.launchdarkly.com/mcp/fm"
    },
    "launchdarkly-ai-configs": {
      "type": "http",
      "url": "https://mcp.launchdarkly.com/mcp/aiconfigs"
    },
    "launchdarkly-observability": {
      "type": "http",
      "url": "https://mcp.launchdarkly.com/mcp/observability"
    }
  }
}

On first use the client opens a browser window to complete the OAuth flow with LaunchDarkly.

Option 2: Local server via npx

Install and run the open source server with a LaunchDarkly API key from the Authorization page:

{
  "mcpServers": {
    "LaunchDarkly": {
      "command": "npx",
      "args": [
        "-y", "--package", "@launchdarkly/mcp-server", "--",
        "mcp", "start",
        "--api-key", "api-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
      ]
    }
  }
}

For Commercial, Federal, or EU instances, pass --server-url with the appropriate LaunchDarkly REST API base URL.

Prerequisites

A LaunchDarkly account
For hosted: a LaunchDarkly user with permission to grant OAuth scopes
For local: an API access token with the scopes needed for the tools you intend to use
Node.js (for the local npx install)

Use Cases

Create a new boolean feature flag, set targeting rules, and roll it out to 10% of users from inside Cursor or Claude Code without leaving the editor.
Audit recent flag changes by listing audit log entries for a specific project and environment.
Manage AI Configs from chat: swap the production model from GPT-4 to Claude on a variation, then update targeting to A/B test it.
Inspect flag health and find stale flags across environments so engineers can clean up technical debt.
Query the observability MCP server for recent error groups or traces tied to a flag rollout to debug regressions.

Example Prompts

"Create a temporary boolean flag called checkout-redesign in the web project and turn it on for 5% of users in production."
"List all feature flags in the mobile project that haven't been evaluated in the last 30 days."
"Show the audit log entries for the payments-v2 flag over the last week."
"Update the AI Config support-bot so the claude-sonnet variation is served to users in the beta segment."
"Query the observability server for error groups in the last hour related to the new-search flag."

Pros

Officially built and maintained by LaunchDarkly under the launchdarkly GitHub org.
Three focused hosted servers cover feature flags, AI Configs, and observability with OAuth (no API key management needed).
Open source MIT-licensed local server is also available for self-hosting and air-gapped environments.
Pairs with LaunchDarkly agent skills for multi-step workflows in Claude Code, Cursor, VS Code, Windsurf, and Codex.

Limitations

Requires a LaunchDarkly account; useful only for teams already on the platform.
Tool surface is split across three hosted endpoints, so users must add multiple MCP entries to access everything.
Local server's tool schema mirrors the REST API and can be verbose for agents (many keys: project, environment, flag, variation).

Alternatives

Statsig MCP for feature flags and experimentation on Statsig.
Unleash MCP community servers for the open source Unleash feature flag platform.
PostHog MCP which includes feature flag management alongside product analytics.