Urgent.News

What's breaking now, across thousands of outlets.

AI

Garry Tan Was Right: "MCP Sucks Honestly." I Have the Token Receipts.

Garry Tan Was Right: "MCP Sucks Honestly." I Have the Token Receipts. "MCP sucks honestly. Context window eats too much, auth is a mess. I wrote a CLI wrapper in 30 minutes and it works better." When YC's CEO says this on X, people listen. But nobody had the data to back it up. Until now. What Garry Tan, Perplexity's CTO, and 97 Million Downloads Can't Hide Three things happened in the last 6…

Garry Tan, CTO of Perplexity AI, admitted MCP is "sucks honestly" after comparing it to a CLI. An independent audit measured the token cost of MCP servers and found the implementation pattern of loading every tool schema into context at startup is a tax on developers. For example, one MCP server listing files on Google Drive injects 47,293 tokens into the context before a single question is asked.

This amounts to 111,713 tokens for just one server, equivalent to a half novel. The audit also revealed that 20% of the context is used for schemas and only 44% is available for reasoning. The cost of this token waste is $2,376 per year using Claude 3.5 Sonnet pricing. The fix is not a new protocol but a smarter proxy layer, such as mcptoon, which reduces schema size by 97%.

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 AI

Claude Code Is Burning Your Token Budget. Here's the Receipt.

Claude Code Is Burning Your Token Budget. Here's the Receipt. I found $2,500/year of hidden token waste in my Claude Code setup. It was the MCP servers.

  • Claude Code using significant token budget, potentially $2,500/year
  • Multiple MCP servers in setup inject schema into every conversation
  • Three-step process to reduce waste: audit servers, remove unused, proxy remaining with mcptoon

MCP Was a Mistake. Here Are 200,000 Tokens That Prove It.

MCP Was a Mistake. Here Are 200,000 Tokens That Prove It. "mcp were a mistake. bash is better." — Peter Steinberger, OpenClaw founder I didn't want to believe it either.

  • MCP protocol touted as AI's USB-C but costly mistake
  • 111,713 tokens injected into context window before interaction
  • 98.7% reduction in token consumption possible with runtime schema loading

More from Sunday 23 August →