|
|
|
|
Every single developer using Claude Code is about to get way more powerful & productive than they already are. |
This new Claude Code update finally fixes a major issue that's been negatively impacting its accuracy for several months now -- and many of us were never even aware. |
All this time, Claude Code has been bloating up your context in the background with unnecessary data from every single one of your MCP servers. |
It didn't matter whether you actually used them or not in any given prompt -- if you have 100 MCP servers, it would dump all the complex tool definitions and metadata for all of them into your context, with no exceptions. |
Drowning out context that actually matters and lowering the accuracy. |
But now with new Tool Search feature in Claude Code, this problem is finally gone forever. |
They've fixed everything -- and they did it in such an amazing way -- web developers would be jealous. |
The old MCP experience was quietly broken |
Here's what was happening before: |
You connect a few MCP servers Each server exposes a bunch of tools Claude loads all of them at startup Your context window gets eaten alive Tool selection gets worse as your tool list grows
|
So even before Claude starts thinking about your actual code, it's already wasting tokens on tool schemas you may never use in that session. |
The more "power user" you became, the worse things got. |
That's backwards. |
Tool Search changes everything -- with a neat trick from web dev |
With Tool Search enabled, Claude Code stops doing dumb work up front. |
Instead of loading everything, it does this: |
Nothing is loaded at startup Claude keeps MCP tools out of context by default When a task comes up, Claude searches for relevant tools Only the tools it actually needs get pulled in Everything else stays out of the way
|
Same MCP. Same tools.
But with lazy loading: Massively better behavior. |
This is exactly how modern AI tooling should work. |
Why this is so huge |
1. You instantly get more usable context |
This is the obvious win — and it matters a lot. |
Tool schemas can be massive. When you're running multiple MCP servers, you're talking thousands (sometimes tens of thousands) of tokens wasted on definitions alone. |
Lazy loading gives that space back to: |
real code repo context actual reasoning
|
That alone makes Claude Code feel noticeably smarter. |
2. Tool selection gets better, not worse |
Too many tools hurt accuracy in another crucial way: |
When a model sees a huge wall of tools, it's harder for it to: |
|
All the stories worth knowing—all in one place. |
|
Business. Tech. Finance. Culture. If it's worth knowing, it's in the Brew. |
Morning Brew's free daily newsletter keeps 4+ million readers in the loop with stories that are smart, quick, and actually fun to read. You'll learn something new every morning — and maybe even flex your brain with one of our crosswords or quizzes while you're at it. |
Get the news that makes you think, laugh, and maybe even brag about how informed you are. |
Try it for free |
AI-native CRM |
|
"When I first opened Attio, I instantly got the feeling this was the next generation of CRM."
— Margaret Shen, Head of GTM at Modal |
Attio is the AI-native CRM for modern teams. With automatic enrichment, call intelligence, AI agents, flexible workflows and more, Attio works for any business and only takes minutes to set up. |
Join industry leaders like Granola, Taskrabbit, Flatfile and more. |
👉Start for free today |
0 Komentar untuk "Claude Code finally fixed its biggest flaw -- this is huge"