RELATED 01
LaunchDarkly local MCP server
OFFICIAL SERVERSelf-run LaunchDarkly server for managing feature flags and AI Configs on EU and Federal instances
27GitHub stars
75k/wknpm downloads
64popularity score
LaunchDarkly hosted MCP server
LaunchDarkly's remote MCP endpoint for feature flags, AgentControl configs and observability data
Official · launchdarkly.com Remote (hosted) · Streamable HTTP Developer toolsDevOps and monitoringAI and ML
Connect your client
Remote (hosted)
Add in one click
Opens the app, which asks before it adds the server.
Or add it by command or config
Cursor.cursor/mcp.json in the project root
{
"mcpServers" : {
"launchdarkly" : {
"url" : "https://mcp.launchdarkly.com/mcp/launchdarkly" ,
"headers" : {}
}
}
}Claude Codeterminal
claude mcp add --transport http "launchdarkly" \
"https://mcp.launchdarkly.com/mcp/launchdarkly"windsurf~/.codeium/windsurf/mcp_config.json
{
"mcpServers" : {
"launchdarkly" : {
"serverUrl" : "https://mcp.launchdarkly.com/mcp/launchdarkly"
}
}
}github copilotrepository Settings > Code and automation > Copilot > Coding agent > MCP configuration
{
"mcpServers" : {
"launchdarkly" : {
"url" : "https://mcp.launchdarkly.com/mcp/launchdarkly" ,
"headers" : {}
}
}
}What it does
This is the MCP server LaunchDarkly runs itself at https://mcp.launchdarkly.com/mcp/launchdarkly. An AI client connects to it over HTTP and signs in to a LaunchDarkly account through OAuth in the browser, so no API access token is created or pasted into a config file. Requests then run with the permissions of the account that authorised the connection.
It covers three product areas. For feature management it can create flags, switch them on or off per environment and change targeting rules. For AgentControl it manages configs. For observability it queries logs, traces, error groups and the Dashboards page. LaunchDarkly's docs also describe a flag removal readiness tool that classifies a flag as safe, caution or not-ready for cleanup, which its agent skills use in a cleanup workflow.
LaunchDarkly recommends this hosted server over its older self-managed npx package (@launchdarkly/mcp-server), which it now keeps for EU and federal instances; its docs say the hosted one is more complete and updated more often, and give migration steps that drop the --api-key argument and LD_ACCESS_TOKEN.
The endpoint's OAuth protected-resource metadata names the MCP endpoint itself as its authorization server, so sign-in is handled by LaunchDarkly at the same address rather than by a separate identity provider.
What it needs
A LaunchDarkly account on the commercial (non-EU, non-federal) instance, with a role that can create, read, update and delete flags and AgentControl configs in the projects used (LaunchDarkly suggests the Writer base role or the Developer preset role). An MCP client that supports remote HTTP servers and OAuth. In Cursor, the server must be toggled on and Connect clicked to finish sign-in.
Limits
Not available on LaunchDarkly's federal or EU instances; those accounts must use the self-managed local server instead. A 403 error after the OAuth consent screen usually means the authorising account lacks permission on the project or environment, not a sign-in failure. LaunchDarkly publishes no tool list for the hosted server on its docs pages; its docs point to the client's own tool list to see what is exposed.
Who it suits
Teams already using LaunchDarkly on the commercial instance who want an assistant to create and change flags, manage AgentControl configs or look at errors and traces without handling API tokens. Because sign-in is OAuth and the code runs on LaunchDarkly's side, nothing is installed locally. EU and federal customers, or anyone who needs to pin a version or run the server on their own machine with an API key, should use LaunchDarkly's self-managed @launchdarkly/mcp-server instead, which LaunchDarkly describes as less complete than this one.