agentskin

AgentSkin Suite

Stop paying for noise. AgentSkin Suite cuts your AI API costs by 60-80% by stripping token waste from API responses, terminal output, and agent replies — automatically.

What You Get

One MCP server. Seven tools. Three layers of savings.

Layer	Cuts	How
API responses	60–88%	Strips junk fields from JSON before your AI sees them
Terminal output	Up to 99.97%	Compacts CLI noise (build logs, git diffs, lint spam)
Agent replies	65%	Makes your AI talk like a caveman — same info, fewer words

30-Second Install

npx -y agentskin@latest

Then add to your MCP config:

{
  "mcpServers": {
    "agentskin-suite": {
      "command": "npx",
      "args": ["-y", "agentskin@latest"]
    }
  }
}

That's it. Restart your AI tool and you have all 7 tools.

What It Actually Does

Before: Your AI fetches https://api.github.com/repos/vercel/next.js and gets 1,544 tokens of JSON — mostly URLs, timestamps, boolean flags you don't need.

After: 180 tokens. Just the fields that matter: name, description, stars, language, topics, URL.

Same answer quality. 88% cheaper API calls.

The 7 Tools

Tool	Use it when...
fetch_optimized_data	You need data from any API or webpage
skin_reasoning	Text is full of hedging and filler words
classify_url	You want to know which rules will apply to a URL
strip_ansi	Terminal output has color codes in it
reduce	CLI output is massive (build logs, diffs, listings)
estimate_tokens	You need to know how many tokens something costs
apply_json_semantic	You have raw JSON you want to prune

Supported AI Tools

Works with anything that supports MCP: Claude Desktop, Claude Code, Cursor, Windsurf, Cline, Codex, Copilot, Kilo Code, OpenCode, and more.

Also includes 6 Caveman skills that work across 30+ agents — compressed output, commits, code reviews, file compression, and subagent delegation.

Numbers

• GitHub API: 88.3% savings (1,544 → 180 tokens)
• Large directory listing: 99.97% savings (3.2M → 897 chars)
• Agent output: 65% smaller
• Test suite: 4,695 tests, 274 files, 100% passing

Docs

• Full Usage Guide — step-by-step setup for every tool
• FAQ — common questions with real benchmarks
• Website — live docs, examples, whitepaper

Security

Everything runs locally. No data leaves your machine. SSRF protection, rate limiting, input validation, and URL sanitization built in.

Credits

Creator	Contribution
Shawn Nichols Sr. (Nichols Transco LLC)	AgentSkin SSS protocol, MCP server, Suite integration
Vincent Koc	Tokenjuice — MIT License
Julius Brussee	Caveman — github.com/JuliusBrussee/caveman

Maintained by Nichols Transco LLC.