FOR CLAUDE CODE make every plan go further

Get more out of Claude. Spend less doing it.

Most of what Claude reads while you debug is junk: repeated log lines, framework stack frames, install spam. UsageGuru cuts it out first, up to 90% fewer tokens, and Claude still sees every error.

On Claude Pro or Max Your usage limits last longer
On the Claude API Your bill gets smaller
1-minute setup Never edits your code Cancel anytime
Works with Claude Code · Codex · Cursor · VS Code · and any MCP tool
Tokens Claude has to read 974

      
server.log, before UsageGuru
−90%tokens on big logs and stack traces
Up to 10×more logs before you hit a limit
$5a month, often less than one day of savings
1 minto set up, then it just works
What it removes

Claude needs the error. Not the 400 lines around it.

Everything you paste stays in the chat and gets re-read on every reply. UsageGuru cuts the parts no one needs, and leaves a marker saying exactly what it cut.

Noise
Removed
Kept
Repeated log lines
382 identical health checks
First and last line, plus [… 382 lines omitted]
Library stack frames
express, node_modules, site-packages
Every frame in your code, with its source line
Duplicate tracebacks
The same crash, printed 9 more times
One full copy, then [same traceback as above]
Install & build spam
npm WARN, pip "already satisfied", docker pulls
The first and last line, so you see start and result
Terminal junk
Colour codes, progress bars, base64 blobs
What the terminal actually showed you
Padded JSON dumps
Indentation and whitespace
Every key and value, compact
−90%
Flask server log
29,919 → 3,033 tokens
−71%
Node dev-server log
974 → 280 tokens
−60%
Failing test run
788 → 315 tokens

Measured with UsageGuru's engine on real-world logs. Bigger, noisier logs shrink more. Lines with different values are never merged, so a single 500 among thousands of 200s stays visible.

How it works

Install once. Your AI does the rest.

1

Add it to your AI tool

Paste one command or config block with the license key you get after checkout. That's the whole setup. Needs Node.js 20+.

$ claude mcp add usage-guru -e USAGEGURU_LICENSE_KEY=your-key -- npx -y usage-guru
2

Claude reads logs through it

When Claude opens a big log or build output, it uses compress_file and gets the short version. To paste something yourself, use trim_prompt.

compress_file("logs/server.log")
[Usage Guru] logs/server.log
Tokens: 29,853 -> 2,969 (saved 26,884, 90.1%)
Removed: 382 repeated lines, 24 library stack
frames, 28 repeated traceback lines, ...
3

See what you saved

Ask Claude for your savings_report: tokens and dollars saved, measured on your machine at real Claude prices.

Usage Guru savings (claude-opus-5-5, $4.00/M)
All time:     214 calls, 3,412,880 tokens saved
Last 30 days:  96 calls, 1,508,302 tokens saved
(example output)
Savings calculator

What is log noise costing you?

Set it to how you work. Uses Anthropic's published API prices.

A 1,000-line log is roughly 15,000 to 30,000 tokens.
Every later reply re-reads it, at a tenth of the price from cache.
Claude model
You'd save about
$0/month
Tokens removed per month0
First read$0
Re-reads$0
UsageGuru costs$5.00
…

Estimate. Assumes 85% of each log is removed and 21 workdays a month; results vary. On a Claude Pro or Max plan you don't pay per token. You hit your usage limits later instead.

Pricing

One plan. $5 a month.

USAGEGURU
monthly plan · cancel anytime
Local log compressionunlimited
compress_file + trim_promptincluded
AI prompt rewrites700/mo
Savings reportincluded
Every Claude modelyes
TOTAL$5/month
Get UsageGuru →
Secure checkout by Stripe · key shown instantly

About one big log a day pays for it.

At Claude's API prices, cutting the noise out of a single 30,000-token log saves around 11 cents, and the same again if it stays in the chat for ten replies. Do that daily and UsageGuru costs you nothing.

27,000 tokens removed
× $4.00 per million (Opus 5.5)
= $0.11 first read + $0.11 re-reads
× 21 workdays = $4.62/month
FAQ

Before you buy.

What exactly is UsageGuru?

A plugin (an MCP server) for AI coding tools, built for Claude Code. It gives your AI tools to read logs, stack traces and command output with the noise removed, so fewer tokens go into your conversation.

Does my code or data leave my machine?

Compression runs on your computer. The only thing ever sent to us is the plain-language part of a prompt you ask trim_prompt to tighten (never code or logs), and you can turn even that off.

Can it hide the error I'm looking for?

It only collapses lines that are truly repeated. Lines with different values stay, errors are never removed, and every cut leaves a marker saying what was removed.

Will it change my code?

Never. Code and source files pass through byte for byte. It also refuses to read .env files, SSH keys and other secrets.

I'm on Claude Pro or Max. Worth it?

You don't pay per token, but you do have usage limits and a context window. Smaller logs mean more work before you hit a limit, and longer sessions before the context fills up.

What happens after 700 AI rewrites?

Log compression keeps working with no limit. Only the optional AI prompt rewrite pauses until the 1st of the month.

What do I need?

Node.js 20 or newer and an AI tool that supports MCP: Claude Code, Codex, Cursor, VS Code (Copilot agent mode), Windsurf, Gemini CLI and others.

How do I cancel or get a refund?

Cancel, change your card or download invoices any time in your billing portal: sign in with the email you paid with. Refunds within 7 days of your first payment, no questions asked. Just open a ticket.

Give Claude the error, not the log.

One command to install. Every log after that is a fraction of the size.

Get UsageGuru for $5/month →