内容
# infy-mcp-playwright
MCP Playwright server for web app testing automation. Exposes an HTTP API to run Playwright tests (navigation, clicks, link checks, screenshots).
This tool can be used for quick website validation, smoke testing, link validation, accessibility testing, performance checks, and basic SEO testing. It reduces the need to manually create a separate Playwright framework for every small testing requirement.
## Install
```bash
npm install infy-mcp-playwright
```
Or install globally:
```bash
npm install -g infy-mcp-playwright
```
## Start the server
**Local project:**
```bash
npm start
```
**Global CLI:**
```bash
infy-mcp-playwright
```
The server prints:
```text
MCP Server Running on Port 3000
```
Set `PORT` to change the listen port (default: `3000`). Set `HEADLESS=false` to open a visible browser window.
## Health check
Verify the server is running without starting a browser test:
```bash
curl http://localhost:3000/health
```
Expected response:
```json
{
"status": "ok",
"service": "infy-mcp-playwright"
}
```
## Run a test request
```bash
curl -X POST http://localhost:3000/run-test \
-H "Content-Type: application/json" \
-d '{
"url": "https://perfectqaservices.com",
"checkLinks": {
"selector": "a"
}
}'
```
### Request body options
| Field | Description |
|-------|-------------|
| `url` | Page URL to open (required) |
| `steps` | Array of user actions to run after the page loads (see below) |
| `type` | `{ selector, value }` — fill an input (legacy single-step shortcut) |
| `press` | Keyboard key (e.g. `"Enter"`) |
| `click` | `{ selector }` — click an element |
| `checkLinks` | `{ selector }` — visit each matching link and report status |
### Multi-step actions
After the page loads, `steps` runs a real user flow. Each step is recorded in `stepResults`. If a step fails, the test stops and returns `status: "failed"` with the failed step.
| `action` | Fields | What it does |
|----------|--------|--------------|
| `click` | `selector`, optional `timeout` | Click an element |
| `type` / `fill` | `selector`, `value`, optional `timeout` | Clear and type into an input |
| `press` | `key` (or `value`), optional `selector` | Press a keyboard key |
| `wait` | `selector` and/or `timeout` | Wait for an element, or wait a number of ms |
| `assertText` | `selector`, `text` | Fail if the element does not contain `text` |
| `screenshot` | optional `name` | Save a full-page screenshot under `artifacts/` |
```bash
curl -X POST http://localhost:3000/run-test \
-H "Content-Type: application/json" \
-d '{
"url": "https://example.com/login",
"steps": [
{ "action": "type", "selector": "#email", "value": "user@test.com" },
{ "action": "type", "selector": "#password", "value": "secret" },
{ "action": "click", "selector": "button[type=submit]" },
{ "action": "wait", "selector": ".dashboard", "timeout": 10000 },
{ "action": "assertText", "selector": "h1", "text": "Welcome" },
{ "action": "screenshot", "name": "after-login" }
]
}'
```
### Programmatic API
```javascript
const { runTest } = require('infy-mcp-playwright');
const result = await runTest({
url: 'https://example.com',
checkLinks: { selector: 'a' }
});
```
## Enhanced checks and outputs
`worker.js` performs additional runtime checks and produces reports and artifacts automatically. Notable request options and outputs:
- **Request body options:**
- `browser`: `chromium|firefox|webkit` — choose the engine (default: `chromium`).
- `headless`: `boolean` — run with a visible browser when `false` (default: `true`).
- `mobile`: `boolean` — emulate `iPhone 13` when `true`.
- `darkMode`: `boolean` — set page color scheme to dark when `true`.
- `slowMo`: `number` — slow motion in ms for Playwright (default: `200`).
- **Runtime checks performed:** performance metrics (dns, tcp, ttfb, dom/load times), memory usage, empty-page detection, infinite-loader detection (common loader selectors), basic SEO extraction (title, meta description, h1, canonical), broken images, console errors, page JS errors, failed requests and network failures (status >= 400), slow API detection (>5s), and accessibility scans via `axe-core`.
- **Link checking:** visits links matched by `checkLinks.selector` (default `a`) with up to 3 retries, captures a screenshot per visited page, and marks each as `PASSED` or `FAILED`.
- **Artifacts & reports:** saved under `artifacts/`, `videos/`, `json-reports/` and `reports/`. The worker writes a JSON report (`json-reports/report-<ts>.json`) and an HTML report (`reports/report-<ts>.html`) and stores screenshots in `artifacts/`.
- **Response object:** the API returns a structured object containing `status`, `browser`, `loadTime`, `performanceMetrics`, `memoryUsage`, `seo`, `totalPages`, `passed`, `failed`, `brokenImages`, `consoleErrors`, `jsErrors`, `failedRequests`, `networkFailures`, `slowApis`, `accessibilityIssues`, `stepResults`, `checkedPages`, `aiSummary`, `htmlReport`, `jsonReport`, and `screenshot` (final screenshot path).
Example curl request using the new options:
```bash
curl -X POST http://localhost:3000/run-test \
-H "Content-Type: application/json" \
-d '{
"url": "https://example.com",
"browser": "firefox",
"headless": false,
"mobile": false,
"darkMode": true,
"slowMo": 100,
"checkLinks": { "selector": "a" }
}'
```
## Publish to npm
1. Log in: `npm login`
2. Dry-run: `npm pack --dry-run`
3. Publish: `npm publish`
## License
ISC
MCP 市场