UsageGuru, a new plugin for AI coding tools, launched on Hacker News today with a specific promise: cut the noise from your logs before they hit the model's context window. By compressing verbose output like stack traces and repeated health checks, the tool claims to reduce token usage by up to 90%, allowing developers to stay within Claude Pro limits longer or reduce API bills significantly.

The Mechanics of Noise Reduction

The tool operates as an MCP server, integrating with Claude Code, Codex, Cursor, and VS Code. It doesn't just summarize; it structurally prunes data. For example, 382 identical health check lines are collapsed into a single marker indicating the omission, while library stack frames from node_modules or site-packages are stripped out, leaving only the developer's actual code paths. Crucially, the system guarantees that unique error values remain visible, ensuring a single 500 error among thousands of 200s isn't lost in the compression.

Pricing and Performance Claims

At $5 a month, UsageGuru positions itself as a break-even investment for heavy users. The company cites benchmarks where a Flask server log dropped from 29,919 to 3,033 tokens, and a Node dev-server log fell from 974 to 280 tokens. For API users, the math is straightforward: removing 27,000 tokens from a single large log saves roughly $0.11 at Opus 5.5 pricing, with additional savings from re-reads in cached contexts. The tool includes a savings report feature that tracks tokens and dollars saved locally.

Privacy and Setup Constraints

Privacy advocates will appreciate that compression runs entirely on the local machine. The only data transmitted is the plain-language prompt when using the optional trim_prompt feature, and users can disable this entirely. Setup requires Node.js 20+ and a simple MCP add command. However, the tool is strictly read-only regarding source code; it never edits your files and explicitly refuses to read sensitive files like .env or SSH keys, acting purely as a filter for terminal output and logs.

Key Takeaways

  • UsageGuru reduces log token usage by up to 90% by removing repeated lines, stack frames, and install spam.
  • The tool costs $5/month and claims to pay for itself with a single large log compression per day for API users.
  • Compression happens locally, ensuring code and logs never leave the machine unless the optional AI prompt rewrite is used.
  • It integrates with major AI coding tools including Claude Code, Codex, Cursor, and VS Code via MCP.

The Bottom Line

For developers bleeding context windows on verbose Node or Python stack traces, this is a smart, low-friction optimization. It’s not a magic bullet for poor debugging practices, but it’s a necessary tool for anyone running Claude Code in production environments.