Run experiments from your AI agent

Connect the LaunchDarkly MCP server and your agent can take an experiment end to end in conversation: create it on a flag or AI config with a hypothesis, metrics, and treatments, start the iteration, reallocate traffic, then read the results and stop with a winner shipped and the reasoning recorded.

An agent can now pull experiment and per-metric results, say whether they're healthy and conclusive yet, explain in plain language what happened, and carry out the decision you make. Agent skills guide it through the right sequence for setup, analysis, and execution, so you get the same procedure every time rather than the right tool calls in whatever order the model happens to pick.

Read LaunchDarkly MCP server (https://launchdarkly.com/docs/home/getting-started/mcp) and browse the LaunchDarkly agent skills (https://github.com/launchdarkly/agent-skills) to learn more.