LaunchDarkly hosted MCP server

AI model by LaunchDarkly

Direct AI assistant access to LaunchDarkly runtime control.

A hosted Model Context Protocol (MCP) server that gives AI assistants like Claude Code and Cursor direct access to LaunchDarkly's feature management and agent control tools. It enables developers to manage production code and AI behavior through natural language prompts.

Model specs

The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.

Launched
2026-11-04

Category: AI Infrastructure

License
Proprietary
Released
2026-11-04 (Updated hosted server)
Protocol
Model Context Protocol (MCP)
Authentication
OAuth
Host URL
https://mcp.launchdarkly.com/mcp
Total Tools
154
  • Direct access for AI assistants like Claude Code and Cursor
  • OAuth-authenticated Model Context Protocol (MCP) interface
  • Granular control via toolsets and write access parameters
  • Enables natural language management of flags and agent configs

Overview

The LaunchDarkly hosted MCP server is a specialized interface that gives AI assistants and agents direct access to LaunchDarkly's runtime control capabilities. It allows developers to manage feature flags, guarded releases, and agent configurations directly from AI-enabled tools like Claude Code and Cursor using the Model Context Protocol. The server is hosted directly by LaunchDarkly at mcp.launchdarkly.com and utilizes OAuth for secure authentication. This enables an "agent-native" workflow where software changes and production controls are managed through natural language prompts within the developer's existing IDE or terminal.

Agent Integration and Control

  • AI Assistant Access: The server provides AI assistants like Claude Code and Cursor with direct access to LaunchDarkly's runtime control capabilities.
  • Natural Language Control: Agents can wrap code in feature flags and configure agent behavior using plain English prompts.
  • Toolsets and Organization: The server organizes its tools into specific toolsets such as Account, Agent Control, Approvals, Experiments, Flags, Guides, Observability, and Releases.
  • Agent-Native Workflow: Users can manage code and agents from wherever they already work, including the terminal or IDE.

Governance and Security

  • OAuth Authentication: The server accepts only OAuth sign-in for secure authentication of AI assistant requests.
  • Granular Tool Control: Users can limit tool access by configuring toolsets and writes URL parameters in the connection URL.
  • Permission Enforcement: The server ensures the strictest evaluation of both the user's LaunchDarkly role and the client-side tool settings.
  • Write Access Management: The server supports additive, destructive, and update writes, which can be toggled in the connection settings.

Automated Workflows

  • Feature Flag Management: The server provides tools to create, update, archive, and manage feature flags directly through the AI assistant.
  • Agent Configuration Tools: AI agents can create AgentControl configurations, variations, and judge-mode evaluations.
  • Approval Workflows: The server includes tools to create, apply, and manage approval requests for flags, segments, and agent configurations.
  • Experimentation Workflows: Agents can set up metrics, instrument code variations, and build experiments through the protocol.

Use cases

  • Developers use the MCP server to ask AI assistants to wrap new code in feature flags using plain English.
  • Engineering teams set up guarded releases and rollback rules directly from their IDE via an AI agent.
  • AI engineers configure and benchmark agent behavior variations without leaving their coding environment.
  • Platform teams restrict AI assistant capabilities to read-only or additive-only writes for specific environments.

Specifications

License
Proprietary
Access
Hosted at https://mcp.launchdarkly.com/mcp for use in MCP-enabled editors and tools
Released
2026-11-04 (Updated hosted server)
Protocol
Model Context Protocol (MCP)
Authentication
OAuth
Host URL
https://mcp.launchdarkly.com/mcp
Total Tools
154
Supported Clients
Cursor, Claude Code, Claude Desktop, VS Code, Antigravity CLI, Codex CLI, opencode, Pi
Tool Groups
Account, Agent Control, Approvals, Experiments, Flags, Guides, Observability, Releases
Write Modes
Additive, Destructive, Update

How it works

  1. Select Client Tool: Select the target AI coding tool such as Cursor, Claude Code, or VS Code from the MCP server page.
  2. Configure Toolsets and Access: Configure the connection by selecting specific toolsets (e.g., Flags, Experiments) and write access levels (Additive, Destructive).
  3. Establish Connection: Update the installation links and add the generated URL to your AI client to establish the connection.
  4. Authenticate via OAuth: Sign in via OAuth to authenticate the AI assistant with your LaunchDarkly account and permissions.

FAQ

Does LaunchDarkly support AI agents?

LaunchDarkly supports agents through the Model Context Protocol (MCP), CLI, and IDE integrations, allowing AI assistants to interact with the platform directly.

Which tools can I use with the LaunchDarkly MCP server?

You can connect to the server from MCP-enabled editors like Cursor, or tools like Claude Code and the Claude Desktop app.

What specific capabilities are available through the MCP server?

The server organizes tools into categories like Account, Agent Control, Approvals, Experiments, Flags, Guides, Observability, and Releases.

View MCP setup

Added by AntSpace from public sources (launchdarkly.com).

More AI models by LaunchDarkly