Urgent.News

What's breaking now, across thousands of outlets.

Tech

Browser MCP token cost: 68 runs of Playwright MCP vs Chrome DevTools MCP, and a proxy that cuts snapshots by up to 96%

You often hear that hooking a browser MCP server up to Claude Code burns tokens, and some posts put it at 10,000 to 50,000 tokens every time the agent looks at a page. I wanted to know whether that holds up, and whether it can be reduced, so I measured the two most popular servers: Microsoft's Playwright MCP and Google's Chrome DevTools MCP. Along the way I ended up building a small tool to cut…

When hooking up a browser MCP server to Claude Code, it can burn tokens, with estimates ranging from 10,000 to 50,000 tokens per page look. To test this, the author measured Microsoft Playwright MCP and Google Chrome DevTools MCP. Using Playwright MCP 0.0.83 and Chrome DevTools MCP 1.11.0, both headless, 68 runs were conducted, totaling 68 MCP runs.

Playwright MCP input tokens for everyday tasks averaged 178,408 tokens ($0.19) and Chrome DevTools MCP input tokens 199,737 tokens ($0.22). The average cost per run was $0.205 for Playwright and $0.236 for Chrome DevTools. Everyday tasks were lighter than expected, with only a small percentage of the total cost.

However, reading long pages in one go can cost over 200,000 tokens, and at OpenAI's o200k tokenizer, this comes out more than 50% larger than Claude's count. Text with many link targets and element IDs is especially expensive for Claude's tokenizer. Fortunately, Claude Code already prevents this by refusing the result as too large and saving it to a file.

A proxy tool, tokensave-mcp-diet, was built to cut the cost by compacting large snapshots, sending only relevant elements, and allowing on-demand access to the full snapshot. Using this proxy on Hacker News and Wikipedia showed an average 63% reduction in tokens for Hacker News and a 96% reduction for Wikipedia, resulting in significant cost savings.

The key takeaway is that most of the browser-MCP bill is the number of steps, not the page size. Asking for specific elements or short scripts can reduce costs, and using a proxy like tokensave-mcp-diet can further minimize expenses.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in Tech

Don't write the prompt. Get interviewed

Everyone teaches you to write prompts. In practice, a person looks at a sample a couple of pages long, with a stack, constraints, formats and edge cases, and closes the tab: there's no way I'll write…

  • Instead of writing prompts, conduct an interview with the agent.
  • Ask targeted questions about system behavior and decision limits.
  • Use interview insights to construct the final prompt.

🦥 TidySloth CLI: The Local-First File Organizer That Embraces Sloth Mode

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady What I Built TidySloth CLI is a lightweight, privacy-focused, local-first Python command-line utility…

  • TidySloth CLI organizes files locally on user's machine
  • Automatically sorts files into subfolders like Code, Documents, Images
  • Emphasizes privacy by processing data offline with no cloud transmission

More from Thursday 22 October →