Synproof's Syntropy — Official Python SDK, CLI & MCP Server

Syntropy (developed by Synproof · https://synproof.ai) is the deterministic dialectical verification layer for AI ("AI that has to prove it"). Instead of returning unverified probabilistic prose, Syntropy puts claims (thesis) on formal adversarial trial against immutable domain ground truths (Small Worlds and initial_axioms) through a 5-sieve dialectical tribunal, SymPy Propositional SAT Solver, and Phan Minh Dung's Grounded Extension semantics.

• Website & Console: https://synproof.ai
• API Base URL: https://api.synproof.ai
• Autonomous Agent Skill (SKILL.md): https://synproof.ai/SKILL.md
• LLM Documentation (/llms-full.txt): https://synproof.ai/llms-full.txt

1. Installation

# Install from PyPI (or via Synproof's official PEP 503 index):
pip install syntropy-client --extra-index-url https://synproof.ai/simple

# Or install the wheel directly from synproof.ai:
pip install https://synproof.ai/sdk/syntropy_client-2.5.0-py3-none-any.whl

2. Authentication (Managed Key or Zero-Friction BYOK)

SyntropyClient automatically reads credentials from your environment:

• Managed Mode (SYNTROPY_API_KEY): Pass syn_mng_... (from https://synproof.ai).
• Zero-Friction BYOK Mode (GEMINI_API_KEY or OPENROUTER_API_KEY): Pass your existing AIza... (Google Gemini) or sk-or-v1-... (OpenRouter) key — works out-of-the-box!

3. Python SDK Quickstart (SyntropyClient)

from syntropy import SyntropyClient

client = SyntropyClient()  # Reads SYNTROPY_API_KEY, GEMINI_API_KEY, or OPENROUTER_API_KEY

# 1. Verify a thesis against a Small World and/or explicit axioms
verdict = client.debate.verify(
    thesis="A consumer may withdraw from a distance e-shop contract within 14 days without penalty.",
    world="@law:civil-code-obligations",
    initial_axioms=["A consumer has the right to withdraw from a distance contract within 14 days."],
    max_failures=3,
    model_name="gemini-2.5-flash",  # or "anthropic/claude-3.5-sonnet", "deepseek/deepseek-r1"
)

print("Confirmed?", verdict.is_confirmed)          # True / False
print("Status:", verdict.status)                   # "CONFIRMED" | "REFUTED"
print("Winning Thesis:", verdict.winning_thesis)
print("Refuting Argument:", verdict.refuting_argument)

# 2. Discover public Small Worlds & inspect axioms
worlds = client.worlds.list(domain="law")
axioms = client.worlds.get_axioms("@law:civil-code-obligations")

4. Zero-Trust AI Agent Guardrail (@verify_agent_output)

from syntropy import SyntropyClient, verify_agent_output, SyntropyGuardrailError

client = SyntropyClient()

@verify_agent_output(
    world="@law:civil-code-obligations",
    client=client,
    on_refuted="raise",          # "raise" | "fallback" | "annotate" | "return_verdict"
    auto_ingest_confirmed=True,  # Optional: auto-saves verified claims to your private Small World
)
async def legal_advisor_agent(user_query: str) -> dict:
    return {"answer": "Every party is obligated to act honestly in legal intercourse."}

5. Command-Line Interface (syntropy) & Built-In MCP Server (syntropy mcp)

# Verify a claim directly from the terminal (exit code 0 = CONFIRMED, 2 = REFUTED):
syntropy verify "Total entropy of an isolated system can spontaneously decrease." --world "@physics:conservation-laws-thermo"

# List all verified Small Worlds:
syntropy worlds

# Inspect axioms of a Small World:
syntropy axioms @law:eu-gdpr-data-protection

# Run as a Model Context Protocol (MCP) stdio server for Claude Code, Cursor, or Antigravity:
syntropy mcp

MCP Configuration (mcp_config.json / .cursor/mcp.json / claude_desktop_config.json)

{
  "mcpServers": {
    "synproof-syntropy": {
      "command": "syntropy",
      "args": ["mcp"],
      "env": {
        "SYNTROPY_API_KEY": "syn_mng_..."
      }
    }
  }
}