| name | playwright-automation |
| description | Use when Codex must browse, interact with, or test web flows via the remote Playwright MCP server hosted on Glama. |
Playwright Automation
Purpose
Drive browser automation workflows (regression checks, screenshot capture, web navigation) through the Glama-hosted Playwright MCP server defined under servers/playwright.
Setup Checklist
- Ensure
GLAMA_PLAYWRIGHT_ENDPOINTandGLAMA_API_KEYare exported and referenced inmcp.json. - Upload authentication state JSON for apps that require login; note the storage path in
servers/playwright/README.md. - Limit concurrency to one session per task to avoid Glama quota throttling.
Workflow
- Intent – describe the scenario (URL, steps, expected assertion). Decide if headless mode is sufficient or if video capture is needed.
- Launch – invoke the
openPage/runStepstool with selectors and navigation instructions. Use deterministic selectors, not text heuristics, whenever possible. - Validate – capture screenshots or DOM snapshots; store outputs under
docs/automation/through the filesystem skill if they must persist. - Cleanup – close tabs via the MCP toolset to release capacity. Record any failures and request a retry only after adjusting selectors/timeouts.
Notes
- Default viewport is 1280×720; override via tool metadata if layout-specific testing is required.
- Retries: wait 3 seconds when encountering
TargetClosedErrorbefore re-issuing the step list. - Sensitive flows (2FA) should be mocked or limited to staging credentials—never push secrets into step definitions.