Skip to main content

Use it with Cursor

This takes about five minutes. You need VegaLoad installed. See Install.

1. Set up the project​

Open a terminal in your project folder. Run:

vegaload init -editor cursor

It writes three files:

created      .cursor/rules/vegaload.mdc
created      .cursor/rules/vegaload-smoke.mdc
created      .cursor/mcp.json

The two rules files tell Cursor's agent how to use VegaLoad. One is for real load tests. One is for a quick ten-second check. .cursor/mcp.json registers the VegaLoad server. It runs vegaload mcp serve.

Run vegaload init with no -editor to set up Cursor and Claude Code together.

note

.cursor/mcp.json holds the full path to the vegaload binary on your machine. If you commit it, a teammate with the binary in another place should run vegaload init -force once. vegaload doctor tells you when the path does not match.

2. Check that it works​

vegaload doctor

For Cursor you should see lines like these:

✓ host.cursor.config      vegaload is registered in .../.cursor/mcp.json
✓ host.cursor.rules       agent rules are present and match this version
✓ host.cursor.handshake   handshake ok: 9 tools (...)

Doctor starts the server, does a real handshake, and expects all nine tools. If a line shows ! or ✗, it says how to fix it. A warning about Claude Code on the same list is fine if you only set up Cursor. It means Claude Code is installed and has no VegaLoad entry in this project. vegaload doctor -fix repairs what it safely can.

In Cursor, open the settings and find the MCP section. vegaload should be listed, with its nine tools. If Cursor was open before you ran init, reload the window.

3. Ask for a test​

Open the agent chat. Ask in plain words:

Write a load test for the checkout API and run it against localhost:8080.

The agent can:

  1. Write a scenario file next to your code.
  2. Check it with one user and one iteration (validate_scenario).
  3. Run it (run_test).
  4. Read the result and tell you what happened.

Localhost is always allowed. For any other host, see Safety. More prompts are in Prompts that work.

4. Read the result​

A run writes a self-contained HTML report. The agent gives you its path. Open it in a browser. It works offline.

To keep the result as a baseline, ask: "Save this report as the baseline." Later, ask: "Run it again and compare with the baseline."

If something is wrong​

  • Cursor does not list vegaload. Run vegaload init -status. It shows each file as ok, missing or outdated. Run vegaload init -force to fix what it reports.
  • The tools list is empty. Run vegaload doctor. It starts the server the same way Cursor does.
  • The agent does not use VegaLoad. Name it in your prompt: "Use VegaLoad to ...".