LaunchDarkly MCP Server – Flags and Observability

LaunchDarkly's official MCP server gives AI agents structured access to feature management, AgentControl, and observability capabilities. Use it to create and update flags, manage targeting, inspect AI configs, and investigate operational signals directly from an MCP-compatible client.

#feature-flags#release-management#observability

Overview

LaunchDarkly's official MCP server connects compatible AI agents to
LaunchDarkly through natural-language tools. It provides current, structured
access to feature-management, AgentControl, and observability workflows instead
of relying only on static model knowledge or manually copied dashboard data.

What the MCP server enables

LaunchDarkly provides hosted MCP capabilities across several product areas.
Depending on the authenticated user's permissions and available tools, an AI
agent can:

Create, inspect, update, and manage feature flags.
Turn flags on or off across environments.
Update targeting rules and rollout behavior.
Review flag-removal readiness and related evidence.
Create and manage AgentControl configs and variations.
Query observability data such as logs, traces, errors, and dashboards.
Use metrics and other supported LaunchDarkly resources.
Combine MCP tools with LaunchDarkly agent skills for multi-step workflows.

When to use it

Use LaunchDarkly MCP when an agent needs live release-control or observability
context. Typical workflows include creating a flag for a new feature, enabling
it in selected environments, changing audience targeting, reviewing whether an
old flag is safe to remove, managing AgentControl configurations, and
investigating recent error groups or telemetry from an AI coding assistant.

Connection and authentication

LaunchDarkly recommends the hosted Streamable HTTP endpoint at
https://mcp.launchdarkly.com/mcp/launchdarkly. The hosted server uses an
interactive OAuth flow and supports feature management, AgentControl configs,
and observability.

LaunchDarkly also publishes the official @launchdarkly/mcp-server package for
local stdio use. The local server requires an API access token with permissions
for the tools being used. LaunchDarkly recommends environment variables instead
of placing tokens directly in MCP client configuration files.

Key considerations

Access is limited by the authenticated user's LaunchDarkly permissions or the
role assigned to the local API token. Use the least privilege required and
restrict the local server with --scope read or repeated --tool options when
appropriate. Review write operations before execution, especially flag state,
targeting, rollout, and AgentControl changes. LaunchDarkly recommends migrating
from the older local npx configuration to the hosted OAuth server for most
clients.

Frequently Asked Questions

When should an AI agent use the LaunchDarkly MCP server?

Use it when a workflow needs current LaunchDarkly context or approved release-control actions, such as creating feature flags, changing targeting rules, managing AgentControl configs, reviewing flag-removal readiness, or investigating logs, traces, errors, and dashboards.

What does the LaunchDarkly MCP server add to an AI agent's capabilities?

It gives the agent structured access to live LaunchDarkly tools and resources, allowing it to manage feature delivery and investigate observability data instead of relying only on static model knowledge or manually copied information.

What can an AI agent access or manage through LaunchDarkly MCP?

Depending on permissions and enabled tools, the agent can work with feature flags, flag states, targeting rules, rollouts, AgentControl configs and variations, flag-removal readiness, metrics, logs, traces, errors, dashboards, and other supported LaunchDarkly resources.

How is authentication configured for the LaunchDarkly MCP server?

The hosted server uses an interactive OAuth flow. The local stdio server requires a LaunchDarkly API access token supplied to the --api-key option. Store the token in a protected environment variable, use least privilege, and avoid placing the real token directly in client configuration files.

Which transport should be used for the LaunchDarkly MCP server?

Use Streamable HTTP with LaunchDarkly's hosted OAuth endpoint for most modern clients. Use the official local stdio package when a local process, custom instance URL, or explicit tool and scope restrictions are required.