loading every MCP server on every prompt was quietly destroying my token budget

via

ClaudeAI

had like 5 or 6 MCP servers configured and did not realize all of them were loading every single time i sent a prompt. even for the dumbest simplest questions. found a routing layer that only loads the relevant ones per prompt and token usage dropped a lot. prompts feel faster too. honestly cannot believe i let it go on that long without checking

More from ClaudeAI

This guy can not be real

ClaudeAI·7

I built /graphify, 26 days, 450k+ downloads, ~40k stars. Here’s what I didn’t expect.

ClaudeAI·5

I dont even have a funny title for this one, I'm just confused

ClaudeAI·3

Whatever SVG/Python based Photo editing claude does is really Good.

ClaudeAI·3

It’s a Weird Time to Be Named Claude

ClaudeAI·3

Claude has a friend?

ClaudeAI·3