Skip to main content

playwright-mcp

Drive a local browser through Microsoft's Playwright MCP server. Use when an agent needs to open a page and act on it.

Quellinformationen

Repository
oidsdev/honeypot
Letzte Quellaktivität
10. Oktober 2026 um 02:46
Erkannte Sprache von SKILL.md
Englisch
Sterne
0
Forks
1

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

Datei-Explorer
2 Dateien

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
playwright-mcp
description
Drive a local browser through Microsoft's Playwright MCP server. Use when an agent needs to open a page and act on it.
# Playwright MCP Open pages and click or type through [Playwright MCP](https://github.com/microsoft/playwright-mcp), Microsoft's MCP server for browser automation (`@playwright/mcp`, Apache-2.0). The server hands back an accessibility snapshot. Actions use element refs from that snapshot. ## When to use A task needs a real page: open a URL, read what is on it, fill a form, or check that a control is there. The MCP client must already be able to start a local stdio server. ## Install Node.js 18 or newer. Confirm the package, then install the browser it launches. ```sh node -v npx -y @playwright/mcp@latest --version npx -y @playwright/mcp@latest install-browser chromium ``` Point the MCP client at the server. `--headless` matters on a machine with no display. Headed is the default. `-y` lets `npx` install the package without a prompt. ```json { "mcpServers": { "playwright": { "command": "npx", "args": ["-y", "@playwright/mcp@latest", "--headless"] } } } ``` ## Commands Call these tools through the MCP client. Names and fields below are the ones the server publishes. 1. `browser_navigate` with `url`. The result includes a snapshot. Each element has a ref. 2. On a large page, `browser_find` with `text` returns matching nodes and their refs. 3. `browser_click` or `browser_type` with `target` set to a ref from that latest snapshot. `browser_type` also takes `text`. `browser_snapshot` captures the tree again. Pass `depth` when the full tree is too large. ## Gotchas - Use a ref from the latest snapshot. A click or navigation that changes the page needs a new snapshot before the next `target`. - `browser_take_screenshot` is for looking. Clicks and typing take a snapshot ref. - Core tools are on by default. `vision`, `pdf`, `devtools`, `network`, `storage`, `testing`, and `config` stay off until you add them with `--caps`. - `browser_run_code_unsafe` runs arbitrary JavaScript in the server process. Leave it alone. - The server binds to localhost. It is not a security boundary. Leave `--host` at the default. - File access stays inside the workspace roots, and `file://` URLs are blocked. Turn on unrestricted file access only when the user asked for that. - A normal run stores a profile under `~/.cache/ms-playwright/` on Linux. Pass `--isolated` when the session should stay in memory. --- Also on Honeypot — a free, open index of skills for AI agents: https://honeypot-e6c.pages.dev [![Indexed on Honeypot](https://honeypot-e6c.pages.dev/badge.svg)](https://honeypot-e6c.pages.dev)
Auf GitHub ansehen