Skip to content
All tags

#browser-automation

5 posts
tech deep-dive

Choosing a Browser MCP: CDP, Playwright MCP, or Puppeteer MCP?

@playwright/mcp uses an accessibility tree instead of screenshots, cutting token cost by 10–50x — the best default for AI agents doing web automation. Puppeteer MCP fits screenshot-heavy tasks. Direct CDP via MCP is for low-level tooling or domains that Playwright/Puppeteer don't expose.

tech deep-dive

Chrome DevTools MCP: An MCP Server Built on CDP

Chrome DevTools MCP wraps Chrome DevTools Protocol (CDP) as an MCP server, giving AI agents direct access to 40+ CDP Domains including Profiler, HeapProfiler, and Security that Playwright and Puppeteer MCP don't expose — at the cost of having to implement MCP tool definitions and auto-wait logic yourself.

tech deep-dive

@playwright/mcp: Microsoft's Official Browser Automation MCP Server

@playwright/mcp defaults to an accessibility tree (browser_snapshot) instead of screenshots, cutting token consumption by 90%+. Combined with Playwright's native auto-wait, it's the best starting point for AI agents doing web automation.

tech deep-dive

@modelcontextprotocol/server-puppeteer: The Official Puppeteer MCP Server

server-puppeteer is the Puppeteer wrapper in the official MCP servers monorepo — seven lean tools built around screenshots and evaluate. Token cost is significantly higher than @playwright/mcp per interaction, but it fits well when the screenshot itself is the deliverable or custom JS execution is the core need.

tech deep-dive

AI-Powered E2E Testing: How canary, Stagehand, Magnitude, and Shortest Each Solve the Problem

AI agents running tests are non-reproducible; hand-written Playwright is hard to maintain. Four tools that emerged in 2024-2025 each tackle this dilemma with very different design philosophies.