Playwright CLI vs Playwright MCP: Which Should You Use with Claude Code?

Short summary

This post compares Playwright MCP vs Playwright CLI for browser automation with Claude Code, with concrete token benchmarks. MCP streams full page state into the conversation (~114K tokens per multi-step task) while CLI writes snapshots to disk and lets Claude grep only what it needs (~26K tokens). MCP is better for sandboxed environments without shell access; CLI wins for CI pipelines and cost-sensitive long sessions.

• •Playwright CLI uses ~26K tokens vs ~114K for MCP on the same multi-step browser task
• •MCP is required for sandboxed clients like Claude Desktop; CLI works anywhere Claude Code runs
• •CLI writes YAML snapshots to disk instead of injecting full page state into model context

Generated with AI, which can make mistakes.