LaunchDarkly MCP Server
Manage feature flags, targeting, and rollouts in LaunchDarkly from your AI agent via MCP.
Updated
What is LaunchDarkly MCP?
The LaunchDarkly MCP server is a Model Context Protocol server that lets AI clients manage LaunchDarkly feature flags, AI Configs, targeting rules, and gradual rollouts through natural language.
LaunchDarkly offers an official hosted server at https://mcp.launchdarkly.com/mcp/launchdarkly, which connects your AI client to LaunchDarkly using OAuth and covers feature management, AgentControl (AI Configs), and observability. LaunchDarkly recommends the hosted option for most users.
A self-hosted/local server is also available as an open-source (MIT) npm package and standalone binary — run it with npx -y --package @launchdarkly/mcp-server -- mcp start --api-key <YOUR_API_KEY>, authenticating with a LaunchDarkly API key instead of OAuth. The server is written in TypeScript and supported clients include Claude Code, Cursor, VS Code with Copilot, and Windsurf.
Tools the LaunchDarkly MCP server exposes
| Tool | What it does |
|---|---|
| FeatureFlags.list | List feature flags in a project. |
| FeatureFlags.get | Get a single feature flag's details. |
| FeatureFlags.create | Create a new feature flag. |
| FeatureFlags.patch | Update a feature flag, including targeting. |
| FeatureFlags.delete | Delete a feature flag. |
| FeatureFlags.getStatus | Get the status of a feature flag in an environment. |
| Environments.listByProject | List environments in a project. |
| AiConfigs.list | List AI Configs (AgentControl). |
| AiConfigs.updateTargeting | Update targeting for an AI Config. |
| AuditLog.listEntries | List audit log entries. |
Connect the LaunchDarkly MCP server
Claude Code (hosted, OAuth)
- 1
Add the hosted server
Run: claude mcp add --transport http "launchdarkly" "https://mcp.launchdarkly.com/mcp/launchdarkly"
- 2
Authorize via OAuth
On first use Claude Code prompts an OAuth login to LaunchDarkly; complete it in your browser to grant access.
- 3
Verify
Run /mcp in Claude Code to confirm the launchdarkly server is connected and its tools are listed.
Cursor (hosted, OAuth)
- 1
Add the server config
Create or edit .cursor/mcp.json and add the LaunchDarkly server URL https://mcp.launchdarkly.com/mcp/launchdarkly.
- 2
Enable it
Open Settings > Cursor Settings > Tools & Integrations > MCP Tools, toggle on LaunchDarkly.
- 3
Connect via OAuth
Click Connect to authorize the LaunchDarkly OAuth flow; the tools become available once connected.
Sample use cases
An agent audits feature flag status across environments on a schedule.
Drift and stale flags surface automatically without manual dashboard checks.
A developer asks their AI client to create or update a flag's targeting rules in natural language.
Flag changes happen inside the editor, governed by LaunchDarkly permissions.
An incident-triggered workflow inspects a flag's targeting and recent audit-log entries.
Responders get fast context on what changed without leaving the agent.
Security & permissions
The recommended hosted server (https://mcp.launchdarkly.com/mcp/launchdarkly) authenticates via OAuth — you authorize the connection through LaunchDarkly, and the agent acts within the permissions of your authorized LaunchDarkly account rather than a static key. The self-hosted/local server instead uses a LaunchDarkly API key passed at startup; treat that key as a secret and scope it to the minimum access needed. In both modes the agent can read and modify feature flags, targeting, and rollouts, so grant it only to trusted workflows and review changes against your audit log.
Using LaunchDarkly MCP with Gamut
In a Gamut workflow, the LaunchDarkly MCP server becomes a tool your agent can call on an event- or schedule-triggered run — Gamut manages the OAuth connection so the agent can read flag status and adjust targeting or rollouts as a step. For example, a scheduled job could check feature flag status across environments and report changes, or an incident-triggered agent could be given the LaunchDarkly tools to inspect a flag's targeting. Gamut handles auth and invocation; the actual changes are still governed by your LaunchDarkly permissions.
Frequently asked questions
What is the LaunchDarkly MCP server?
It is LaunchDarkly's official Model Context Protocol server that lets AI clients manage feature flags, targeting, rollouts, and AI Configs in LaunchDarkly using natural language. A hosted server is available at https://mcp.launchdarkly.com/mcp/launchdarkly, plus an open-source local option.
Is the LaunchDarkly MCP server free?
Yes — the MCP server itself is free and open source under the MIT license. You still need a LaunchDarkly account to use it, and that follows your existing LaunchDarkly plan and pricing.
How do I connect the LaunchDarkly MCP server to Claude Code or Cursor?
For Claude Code, run claude mcp add --transport http "launchdarkly" "https://mcp.launchdarkly.com/mcp/launchdarkly" and complete the OAuth login. For Cursor, add the same URL to .cursor/mcp.json, enable LaunchDarkly under Settings > MCP Tools, then click Connect to authorize via OAuth.
Is the LaunchDarkly MCP server hosted or self-hosted?
Both. LaunchDarkly recommends the hosted server (OAuth) at https://mcp.launchdarkly.com/mcp/launchdarkly, and also offers a self-hosted/local server via the @launchdarkly/mcp-server npm package or standalone binary that authenticates with a LaunchDarkly API key.