What is Edgee Claude Code Compressor V2?

Edgee Claude Code Compressor V2 is a token compression tool designed specifically for coding agents. It reduces the number of tokens sent to large language models during coding sessions by applying three distinct compression techniques across two layers. The result is a −50% reduction in tokens on a typical session — dropping from 18,420 to 9,210 tokens — while remaining semantically lossless on code-oriented tasks. It works as a drop-in CLI wrapper with zero code changes and integrates with existing API keys and plans.

Who it's for

• Coding agent users who want to cut token bills without sacrificing code quality or model performance
• Developers using Claude Code, Codex, OpenCode, or Cursor who need a simple, drop-in solution that works with their existing workflows
• Teams running frequent coding-agent sessions who want to reduce costs at scale while keeping the same context and output quality

Key features

Tool result trimming (Layer 1 — Input)

This improved technique filters CLI and tool results before they reach the model. It removes boilerplate, pagination markers, ANSI escape sequences, repeated headers, and verbose framing. A 980-token directory listing becomes a dense 340-token one that the model reads just as well, saving −10% of tokens in a typical session.

Tool surface reduction (Layer 1 — Input)

New in V2, this technique runs a small, fast classifier that scores each tool against the classified task. It then strips or down-scopes irrelevant tools before the request hits the model. Your IDE still exposes everything, but the model only sees a curated, task-relevant subset — no more manual toggling of MCP servers.

Output brevity (Layer 2 — Output)

This new technique reduces the verbosity of model responses without losing technical content. You pick the level (light, medium, hard) to trade aggressiveness against tone. Though output is only ~1% of token volume, it's the traffic you pay the most for, making this −30% reduction highly valuable.

What stands out

Compression V2 cuts token bills by half while keeping every bit of context the model needs — semantically lossless on code tasks.

This isn't just about trimming tokens. The three techniques work together across both input and output layers, each independently toggleable via config flags. The tool surface reduction is particularly clever — it eliminates the need to manually manage which MCP servers are active, letting the classifier handle relevance automatically. With <12ms P50 gateway overhead and zero code changes, the performance impact is negligible while the cost savings are substantial.

Worth checking out if…

You use coding agents like Claude Code, Codex, OpenCode, or Cursor and want to reduce token costs without changing your workflow. It's especially valuable if you run frequent sessions where even small per-request savings add up quickly, or if you're tired of manually toggling MCP servers on and off. The drop-in CLI wrapper means you can start saving tokens in minutes with no code changes.