@tokenectomy/sentinel

Stop wasting 50,000 tokens every time your app crashes in Claude & Cursor.

When a Next.js, Node, or Django app throws an error, your terminal prints hundreds of lines of internal library stack traces (node_modules, site-packages, internal runtime traces).

When you paste that into Claude or let Cursor read it, you burn 30,000 to 60,000 tokens on library code you never wrote. The AI gets confused, loses context, and hallucinates broken fixes.

Tokenectomy Sentinel runs locally on your machine as an MCP server. In less than 1 millisecond, it cuts out the 95% framework junk and redacts leaked passwords or API keys—giving your AI agent only the exact lines of your code that caused the crash.

📊 Real Token Economics (GPT-4o Benchmark)

Measured using OpenAI's cl100k_base tokenizer across common development crashes:

| Crash Scenario | Raw Error Size | With Tokenectomy | Tokens Cut | Cost Saved / Error | |---|---|---|---|---| | Next.js Hydration Crash | 58,300 tokens | 74 tokens | -99% | $0.29 | | Node.js Express Crash | 38,000 tokens | 62 tokens | -99% | $0.19 | | Python Django REST API | 40,500 tokens | 10,500 tokens | -74% | $0.15 | | Python PyTorch / ML | 43,800 tokens | 8,800 tokens | -79% | $0.17 | | Rust Tokio Runtime Panic | 42,700 tokens | 22,300 tokens | -47% | $0.10 |

Typical Team Impact: A team generating 2,000 errors/month saves ~72,000,000 tokens (~$360/month or $4,350/year) in raw LLM API spend.

🔬 Test It On Your Own Terminal (5 Seconds, Free)

Run the benchmark on your own computer without installing anything:

npx -y @tokenectomy/sentinel --benchmark

🔌 Setup in Cursor & Claude Desktop (1 Minute)

1. Cursor

Add to .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "tokenectomy": {
      "command": "npx",
      "args": ["-y", "@tokenectomy/sentinel", "--mcp"]
    }
  }
}

2. Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "tokenectomy": {
      "command": "npx",
      "args": ["-y", "@tokenectomy/sentinel", "--mcp"]
    }
  }
}

3. Google Antigravity CLI

agy mcp add tokenectomy -- npx -y @tokenectomy/sentinel --mcp

🛡️ What Tokenectomy Does Behind the Scenes

1. Log Surgery: Strips out framework callstacks (node_modules, venv, site-packages, internal runtime) in <0.2ms before the AI agent reads them.
2. Secret Redaction: Automatically blacks out JWTs, database passwords, and API keys before they leak to external LLMs.
3. Blast Radius Protection: Warns the agent before it modifies widely-used functions across multiple files so it doesn't silently break other parts of your app.
4. Time Machine Rollback: Reverts broken AI patches with a single command if the agent hallucinates:

   npx @tokenectomy/sentinel --undo

📦 License & Pro Features

• Free Core: CLI benchmark, standalone error log sanitizer, and core MCP scrubbing.
• Pro Tier: Unlocks automated multi-file blast radius defense, AST anti-hallucination guard, and time-machine rollback:

  npx -y @tokenectomy/sentinel --activate <YOUR_KEY>

  Get a lifetime key: https://tokenectomy.gumroad.com/l/kiznsu

💻 Supported Platforms

• Linux (x86_64)
• macOS Apple Silicon (ARM64)
• Windows (x86_64)