token-optimizer

60–75% token reduction for AI coding agents — one command to install, works everywhere.

What it does

token-optimizer keeps useful information and removes low-value output from agent workflows:

Compresses shell output from git, tests, builds, package managers, grep, Docker, and Kubernetes.
Compacts file reads, web fetches, grep/glob results, and sub-agent task results.
Omits gitignored files from glob, grep, and directory-read outputs.
Encourages concise agent replies to reduce response-token overhead.
Preserves exit codes and falls back to raw output on errors.

Typical savings:

Workflow	Without optimizer	With optimizer	Typical saving
`git status` / `git diff` checks	1,000 tokens	150-300 tokens	70-85%
Test/build logs	4,000 tokens	800-1,500 tokens	60-80%
File search / grep / glob	2,000 tokens	400-900 tokens	55-80%
Long file or web reads	8,000 tokens	2,000-4,000 tokens	50-75%
Basic coding task with several tool calls	12,000 tokens	4,000-6,000 tokens	50-65%

Install

Choose one:

A. npm / npx

npx token-optimizer install

B. curl

curl -fsSL https://raw.githubusercontent.com/edisonaugusthy/token-optimizer/main/install.sh | bash

Both options detect supported agents and patch their config automatically.

Supported agents:

Agent	Integration
OpenCode	MCP server plus plugin hooks
Cursor	MCP server
Claude Desktop	MCP server
Windsurf	MCP server
Codex	`AGENTS.md` shell-filter instructions

Commands

npx token-optimizer status     # Install status and token totals
npx token-optimizer stats      # Token savings summary
npx token-optimizer update     # Pull latest version and reinstall
npx token-optimizer uninstall  # Remove injected configs

Development

git clone https://github.com/edisonaugusthy/token-optimizer
cd token-optimizer
npm install
npm run build
node dist/scripts/setup.js install

Useful scripts:

Command	Description
`npm run build`	Compile TypeScript
`npm start`	Start the MCP server

License

MIT