You’re building fast.
Know what actually works.
Building a web app with Claude Code, Cursor, or Codex? Give your coding agent the tools to explore your app, test important user flows, and catch regressions before you ship.
No account. No telemetry. Tests you keep.
Works with Claude Code, Cursor, Copilot, Codex, and Claude Desktop over MCP.
uv tool install ordel-cli. Install the free CLI (or: pipx install ordel-cli).ordel init. Set up Ordel in your app’s project folder.ordel doctor. Check Node, Playwright, and the browser before testing.ordel mcp-install --client claude-code --write. Restart Claude Code and approve the project server to connect.
Demo only · Requires uv or pipx, Node 18+, and Playwright. Full setup below.
You ask
“Explore my app and test the checkout flow.”
Explore
Map pages, elements, and user journeys.
Test
Turn a flow into real Playwright tests.
Get results
Run in a browser. See what passes and fails.
import { test, expect } from '@playwright/test';
test('a signed-in shopper can place an order', async ({ page }) => {
await page.goto('/cart');
await page.getByTestId('checkout-btn').click();
await page.getByLabel('Card number').fill(process.env.ORDEL_SECRET_CARD ?? '');
await page.getByRole('button', { name: 'Place order' }).click();
await expect(page).toHaveURL(/\/orders\//);
await expect(page.getByText('Order confirmed')).toBeVisible();
});Real Playwright .spec.ts files. In your project. Yours to run.
Put Ordel to work.
Install the CLI, connect your coding agent, and start testing your app.
Install uv
uv installs Ordel and provisions the right Python for you.
Or with Homebrew: brew install uv
Add Playwright to the project you test
Ordel maps, records and runs through a real browser: the Playwright in your project (Node 18+). Run these in that project's folder.
Install Ordel and set up the project
Same folder. init creates .ordel/ and ORDEL.md, like git init; doctor checks Node, npx, Playwright and a real Chromium launch, and says how to fix anything missing.
Prefer pipx? pipx install ordel-cli works the same. No install? Prefix any command with uvx --from ordel-cli to run it on demand.
Connect your coding agent
One command writes or prints the MCP config for your client. The client only sees Ordel after the step below it.
- Writes .mcp.json in this project.
- Restart Claude Code in the project and approve the ordel-cli project server when it asks (or run /mcp and enable it).
Then ask your agent: “use ordel to explore and test the app at localhost:3000.”
What you get
31 local, deterministic tools your agent can call: the camera and the QA brain:
It emits real Playwright .spec.ts you keep and run, no lock-in. Deterministic, with no Ordel LLM: your agent is the brain, Ordel is the truth.
What leaves your machine
- Nothing from Ordel. No account, no telemetry, no Ordel model. Your map, flows and runs stay in the project's
.ordel/folder, and a password typed during a recording is never written to disk. - What your coding agent reads goes to its model. Ordel's results, like page text and test output, reach your agent's own model provider the same way your code does.
- Downloads you start. uv and npm fetch the packages and the browser, and the browser visits the app you point it at.
Free CLI, hosted MCP or local runner?
Three different things. Only the free CLI works without an account.
Ordel's local engine on your machine. Your coding agent drives it; state stays in your project's .ordel/ folder.
This page
Your coding agent works on your Ordel workspace in the cloud: its projects, test cases, runs and history.
A machine that runs your Ordel workspace's tests, such as your laptop or a CI box, dispatched from the cloud.