Automated E2E Testing Engine

Ghostly tests for you

Describe the goal. Ghostly navigates, interacts, and alerts you when something breaks.

npm install -g @ghostly-io/cli

ghostly-cli

$ ghostly up

✓ Local engine started on localhost:4000

✓ Browser ready

✓ Agent waiting for instructions

$ New run: "Log in and reach the dashboard"

● Running plan...

Auto-healed tests

Faster than manual E2E

Zero

Test code written

Local-first

Execution modes

Three ways to drive it

Same engine. Pick the level of control you want.

Assisted

No code

Describe the outcome in plain English; Ghostly plans and runs it. For QA, PMs, and anyone who doesn't write code.

Advanced

For engineers

Hand-write the exact steps as JSON when you want precise, deterministic control over every action.

MCP

In your IDE

Call Ghostly's tools from Cursor, Claude Desktop, or any MCP client — generate, run, and save flows without leaving your editor.

Features

Real browser flows, AI-assisted

Keep shipping fast with full regression guarantees

Assisted Mode v2

The AI analyzes the page (recon), proposes a step plan, and runs it horizon by horizon

Self-healing

If a step fails due to a broken selector, the Healer analyzes and repairs it automatically

Victory conditions

Define success conditions: expected text, visible selector, or resulting URL

Adaptive memory

Reuses previous successful plans to speed up repeated runs

Local dashboard

React UI with projects, run history, coverage metrics, and step-by-step detail

MCP integration

Cursor, Claude Desktop, or any MCP client — tools right in your IDE chat: run_flow, submit_plan, get_project_map

Local-first / Zero-trust

Everything runs on your machine. Your code never leaves your environment.

Screenshots + video

Automatic screenshot on every step; video recorded on failed runs

Self-update

When a new version ships, Ghostly keeps itself current — no reinstall.

Bring your own model

Your model. Your keys. Your machine.

Ghostly never ships an AI. Point it at any OpenAI-compatible endpoint or a local CLI (e.g. Cursor CLI). Your keys stay on your machine.

Any OpenAI-compatible endpoint

Or a local CLI (e.g. Cursor CLI)

Keys never leave your machine

How it works

From zero to testing in 3 steps

No test code to write. No complex setup.

Step 01

npm install -g @ghostly-io/cli
ghostly install   # set up credentials
ghostly up        # start the local engine

Let Ghostly work for you

Install Ghostly in under 5 minutes and start automating your E2E tests with AI.