SYM Protocol

Symbolic prompt compression for LLMs — reduce input tokens 40-65% without losing meaning.

"Necesito que analices el código y corrijas los errores de autenticación en el servidor"
→ "NEED >AN {COD} & >FIX {ERR} {AUTH} {SRV}"   (52% fewer tokens)

The LLM receives a compact system prompt with only the symbols used, understands perfectly, and responds normally.

How it works

Your prompt → [sym compress] → compressed prompt + dynamic dict → LLM → response

1. Replaces known terms with symbols (función→{FN}, refactorizar→>RF, necesito→NEED)
2. Injects only the symbols used into the system prompt (~50 tokens vs 800 static)
3. LLM understands 100% — no training required, symbols are self-explanatory

Install

npm install -g sym-protocol

Use as MCP server (Claude, Cursor, Windsurf, VS Code, Zed...)

sym-install          # auto-detects and configures all supported MCP clients
sym-install --status # check which clients are configured

Or add manually to your MCP config:

{
  "mcpServers": {
    "sym-protocol": {
      "command": "sym-mcp"
    }
  }
}

MCP Tools available

| Tool | Description | |------|-------------| | sym_compress | Compress text → symbolic notation (40-65% token reduction) | | sym_decompress | Expand symbols → natural text | | sym_stats | Compression metrics for a text | | sym_send | Compress + call LLM + return response |

Use as Node.js module

const { compress, decompress, metrics } = require('sym-protocol');

const { compressed, ratio } = compress('Necesito refactorizar la lógica de autenticación');
// compressed: "NEED >RF {LOG} {AUTH}"
// ratio: 55

const expanded = decompress('NEED >RF {LOG} {AUTH}');
// expanded: "need refactor logic auth"

Dynamic system prompt

const { buildDynamicPrompt } = require('sym-protocol/core/sym-prompt');

const prompt = buildDynamicPrompt('NEED >AN {COD} & >FIX {ERR} {AUTH}');
// Returns only the definitions for the 5 symbols used (~50 tokens)
// instead of the full 1,231-symbol dictionary (~800 tokens)

Coverage

1,231 symbols across 38 domains:

• Core dev: functions, APIs, databases, errors, auth, config, tests
• DevOps: Docker, Kubernetes, CI/CD, monitoring, cloud (AWS/GCP/Azure)
• Languages: TypeScript, JavaScript, Python, Rust, Go, Java, React...
• AI/ML: embeddings, RAG, agents, fine-tuning, prompt engineering
• Conversational: questions (HOW?, WHAT?, WHY?), verbs (NEED, WANT, CAN, MUST), connectors (BUT, ALSO, THEN, LAST)
• Business: invoices, inventory, sales, KPIs, meetings, proposals
• Medical, Legal, Finance, Marketing

Benchmarks

| Prompt type | Token reduction | |-------------|----------------| | Technical (code/infra) | 45-65% | | Conversational + technical | 35-52% | | Pure conversational | 25-40% |

CLI

sym compress "Necesito que refactorices el código de autenticación"
sym decompress "NEED >RF {COD} {AUTH}"
sym stats "Tu texto aquí"

Requirements

• Node.js >= 18
• No external dependencies for core compression
• @anthropic-ai/sdk only needed for sym_send tool

License

MIT