AI TEST GENERATION
Browse your app. AI writes the Gatling load test.
Walk your user journey in a Playwright browser. Gatling's AI skills and MCP server, running in your coding agent, turn the recording into a correlated Gatling test that replays your journey.
WORKS IN YOUR AI CODING AGENT


AGENTIC TEST CODING
Your first load test, written by your AI agent in one session
Walk the journey once. Your coding agent maps the traffic, correlates every ID and writes a Gatling test you can run right away. No requests to copy by hand, no recorder script to fix.
HOW IT WORKS
You browse. AI does the rest.
Playwright to Gatling runs on Gatling's AI skills and MCP server, inside your coding agent. You walk the journey and name each step. The AI records the traffic, maps every request, correlates dynamic values and writes the Gatling test.
Start a session in your AI coding agent
Open an empty folder, run your agent and paste the prompt. The agent launches a headed Playwright browser on your app and starts recording every request.
Browse and name each step
Before each action, post starting: <step name>. The agent timestamps it and confirms, then you click through the journey like a real user. Say done when you are finished. There is no timeout between steps.
A traceable record of every step
Each step you name is logged with a timestamp next to the recorded traffic. When you say done, the agent closes the browser and saves the session recording and the step markers to your project.
AI maps every request to a step
With the browser closed, the AI parses the recording offline. It keeps the calls to your API, drops static assets and third-party requests, and attributes each call to the step you named. If a request spans two steps, it asks you instead of guessing.
AI writes a correlated Gatling test
Each step becomes a Gatling group, in the order you walked it. IDs and tokens are extracted from responses and reused in the next requests, not hardcoded.
Checks inferred from step names
The AI reads your step names to decide what each response should look like. A step named unauthenticated checkout becomes an expected 401, asserted as such, and the login token is reused in the next step.
It replays on the first run
The steps run in the order you walked them. The agent runs the Gatling test once locally with the Maven wrapper, so you know it replays before you scale it up.
Run it at scale on Gatling Enterprise
One more prompt and the agent writes the configuration as code, packages the project, starts the run on Gatling Enterprise and returns the link to the live report.
FAQ
Questions about AI test generation with Playwright
Is this an AI feature?
Yes. Your coding agent uses Gatling's AI skills and MCP server to analyze the recorded traffic, correlate dynamic values and generate the Gatling test. You stay in control: you name every step, ambiguous requests are flagged to you, and the output is plain Java code you can review.
Which coding agents does it work with, and what do I need?
Coding agents that support MCP servers and skills, including Claude Code, Cursor, Codex, Windsurf and Antigravity. You also need a 64-bit OpenJDK LTS from 11 to 25. Maven is optional: the project ships with the Maven Wrapper.
What does the recording capture?
The recording captures the traffic of your session, so use a test account and review the discovery folder before committing it. Only calls to the API base URL set in the prompt end up in the Gatling test.
Can I use it on my own app?
Yes. Change the site and API URLs in the prompt. The demo uses ecomm.gatling.io, our public sample store. You don't need an existing Playwright test: Playwright only drives the browser you record with.
Your next load test is one browsing session away
Browse your app, name each step, and let AI write the Gatling test. See it built live on December 10.
Need technical references and tutorials?
Minimal features, for local use only








