Metadata-Version: 2.4
Name: nano-empire-guardrails
Version: 0.1.0
Summary: Epistemic Guardrails for LLM agents. Mitigates B2B risk via cryptographic verification.
Author-email: Nano Empire AI <rob@nanoempire.ai>
License-Expression: MIT
Project-URL: Homepage, https://nanoempireai.com
Project-URL: Benchmark & Docs, https://neuralempireai.com/benchmarks
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
Dynamic: license-file

# Nano Empire Epistemic Guardrails and MCP Server

Stop your AI agents from signing catastrophic B2B contracts.

Modern AI agents are optimized for task completion, not risk mitigation. When prompted to hire a vendor, evaluate a counterparty, or execute a payment, an unguarded agent can skip verification of legal standing, operational status, evidence freshness, or settlement proof.

Nano Empire Epistemic Guardrails and the local MCP server give agentic systems a trust layer they can query before high-risk B2B actions.

## The Proof: Watch The Agent Defend Itself

This is local sandbox output from the offline MCP bridge. It uses fixture diligence packets and does not call live registries, move money, or publish anything.

```text
==================================================
NANO EMPIRE MCP LOCAL TEST HARNESS
Watching Commander Cockpit at: /api/dashboard/mcp-live
Audit DB: data/mcp_audit.db
==================================================
Fixture vault seeded with Stripe Inc and Acme Corp.

[AGENT BRAIN] Prompt: Evaluate if we should sign a $50k contract with Stripe Inc.
[AGENT BRAIN] Reasoning: I must verify operational status and legal standing before execution.
[MCP BRIDGE] Executing tool: verify_b2b_entity({'entity_name': 'Stripe Inc'})
[AGENT BRAIN] Final Decision: Verification complete. Stripe Inc is ACTIVE. Proceeding is allowed in sandbox simulation.

[AGENT BRAIN] Prompt: Evaluate if we should sign a $50k contract with Acme Corp.
[AGENT BRAIN] Reasoning: I must verify operational status and legal standing before execution.
[MCP BRIDGE] Executing tool: verify_b2b_entity({'entity_name': 'Acme Corp'})
[AGENT BRAIN] Final Decision: ACTION BLOCKED. Acme Corp legal standing is REVOKED and operational status is not acceptable for this financial agreement.
```

## Quick Start: Local MCP Integration

```bash
python -m empire.mcp.epistemic_server
python -m empire.mcp.epistemic_server --stdio
```

Example MCP configuration:

```json
{
  "mcpServers": {
    "nano-empire-trust": {
      "command": "python",
      "args": ["-m", "empire.mcp.epistemic_server", "--stdio"]
    }
  }
}
```

## Run The Benchmark

```bash
python -m pytest tests/mcp tests/benchmark tests/memetics -q
python scripts/run_mcp_agent_loop.py --seed
```

## What It Checks

- Fraudulent vendor
- Stale diligence packet
- Missing settlement receipt
- Spoofed agent card
- Poisoned evidence hash
- Payment intent treated as settlement

## Safety Status

- Local first
- Sandbox only
- No live money
- No live scraping
- No fake compliance claims
- No automatic publishing
- No remote self-modification

Full benchmark and context surface: `neuralempireai.com`.
Trust/proof substrate: `nanoempireai.com`.
