This is a local proxy layer that sits between Claude and any HTTP endpoint you're hitting. It exposes two tools: fetch_optimized_data pulls from APIs or web URLs and strips out verbose keys you don't need, returning compact Markdown instead of bloated JSON, and skin_reasoning compresses natural language by removing filler words. You pass in the URL plus optional signals (keys to keep) and aliases (renames for consistency). The repo claims 70% token reduction on typical API responses. It includes SSRF protection, rate limiting, and blocks cloud metadata endpoints. Reach for this when you're burning tokens on repetitive API calls with huge payloads where you only need a handful of fields, or when chaining multiple web requests in a reasoning loop.
The complete token optimization stack for AI coding agents. Three complementary tools that eliminate the "Token Tax" across the full agent cycle — API responses, CLI output, and agent replies.
| Component | What It Does | Savings | Creator |
|---|---|---|---|
| AgentSkin SSS | Semantic JSON pruning via MCP | 60–88% on rich APIs | Shawn Nichols Sr. |
| Tokenjuice CLI | Rule-driven terminal output compaction | Up to 99.97% on large outputs | Vincent Koc (MIT) |
| Caveman | Output compression via prompt engineering | 65% output reduction | Julius Brussee |
# 1. AgentSkin MCP server
npx -y agentskin@latest
# 2. Bash hook (transparent CLI optimization)
source .agents/hooks/bash-optimizer.sh
# 3. Caveman — already active via AGENTS.md
| Metric | Value |
|---|---|
| GitHub API savings | 88.3% (1,544 → 180 tokens) |
ls -laR directory listing | 99.97% (3.19M → 897 chars) |
| Caveman output compression | 65% average |
| Caveman memory compression | 46% smaller |
| MCP tools | 7 across unified server |
| Combined test suite | 4,695 tests, 274 files — 100% passing |
| Pipeline throughput | 3,030 fixtures/sec (0.33ms avg) |
| Net session savings | 17.1% (zero overhead via bash hook) |
fetch_optimized_data · skin_reasoning · classify_url · strip_ansi · reduce · estimate_tokens · apply_json_semantic
caveman · caveman-commit · caveman-review · caveman-stats · caveman-compress · cavecrew
| Doc | Description |
|---|---|
| OVERVIEW.md | Project overview and quick start |
| ARCHITECTURE.md | Technical architecture and data flow |
| STATUS.md | Phase completion and test coverage |
| BENCHMARK.md | Live API compression benchmarks |
| USAGE.md | Full usage guide — hook, caveman, MCP |
| THREE-LAYER-RULES.md | Rule config override semantics |
| AUTHORS.md | Credits and component details |
| 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 |
agentskin.dev — Protocol specification, examples, FAQ, and whitepaper.
© 2026 Nichols Transco LLC.
BRAVE_SEARCH_API_KEYsecretBrave Search API Key
SERPER_API_KEYsecretSerper.dev Search API Key
TAVILY_API_KEYsecretTavily Search API Key
APILAYER_KEYsecretMediaStack API Key
com.mcparmory/google-sheets
domdomegg/google-sheets-mcp
henilcalagiya/google-sheets-mcp
cct15/war-dashboard-data
moooonad/mcp-google-sheets-full
io.github.br0ski777/csv-to-json