Changelog 2026.08

Connect AI Agents to Loadster

We’re happy to announce the Loadster MCP Server, so AI assistants like Claude, ChatGPT, Cursor, and Codex can work in your Loadster projects directly.

Describe the flow you want to test in plain language, and your agent can build it for you! Your agent can help you with writing and playing scripts of any type (Browser Bots, Playwright, or Protocol Bots), debugging them by actually running them and reading the results, wiring up datasets, assembling load test scenarios, setting up monitors, and reviewing finished test reports to offer advice. The Loadster MCP uses the same building blocks you’d use in the dashboard, so the scripts it produces are ordinary Loadster scripts you can then open and edit yourself.

For safety reasons, your agents can’t spend your fuel on their own. When it’s time to actually launch a load test or start a monitor, the agent hands you a link so you can push the button. Generate an access token from your account settings to get started! The AI Agents page walks through how to connect each assistant.

Other Improvements & Bug Fixes

• Broken dataset variables now show up so you can actually fix them, instead of the editor falling back to an empty state that gave you nothing to repair.
• Error messages from Code Block functions like waitFor are much clearer about what went wrong and how to fix it.
• Command-clicking (or Ctrl-clicking) a link-like element in the dashboard now reliably opens it in a new tab.
• Errors thrown from Code Blocks no longer show up twice in traces.
• Custom browser launch arguments are now merged with Loadster’s defaults instead of replacing them.
• Fixed a case where a monitor could falsely warn that it had no valid monitoring script after an edit.
• Fixed some uncaught errors that could break parts of the test report view.
• Sped up blob storage and retrieval for snappier script editing.