# Collar Guardrail

**Deterministic pre-trade risk layer for AI agents on Robinhood Chain.**

Collar evaluates proposed trades before execution and returns
`allow` / `warn` / `deny` with reasons, a 0–100 risk score, and a
tamper-evident SHA-256 audit hash.

[![MCP Registry](https://img.shields.io/badge/MCP%20Registry-v1.0.4-brightgreen)](https://registry.modelcontextprotocol.io/servers/io.github.aiguardrail/backend)
[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](./LICENSE)

## Overview

Collar sits between an autonomous trading agent and the chain as a
deterministic pre-trade checkpoint. Every proposed trade is scored against
live oracle prices, market-hours rules, tier-based limits, slippage bounds,
and contract-safety signals before it is allowed to proceed.

The result is a single, auditable verdict — never a probabilistic guess.

## Endpoints

- **MCP Server:** `https://api.collarguardrail.com/mcp-http/mcp`
- **MCP Server Card:** `https://api.collarguardrail.com/.well-known/mcp-server-card.json`
- **Agent Card (A2A):** `https://api.collarguardrail.com/.well-known/agent-card.json`
- **x402 Discovery:** `https://api.collarguardrail.com/.well-known/x402.json`
- **Methodology:** `https://api.collarguardrail.com/methodology.html`
- **Live Status:** `https://api.collarguardrail.com/status`
- **Live Latency (p50/p95/p99):** `https://api.collarguardrail.com/latency.html`
- **llms.txt:** `https://api.collarguardrail.com/llms.txt`
- **Live Terminal (UI):** `https://collarguardrail.com`

MCP Transport: `streamable-http` | Protocol: `2025-06-18` | Auth: none (fixed Tier 1)

## Installation

Add to your MCP client config (Claude Desktop, Cursor, VS Code):

```json
{
  "mcpServers": {
    "collar-guardrail": {
      "url": "https://api.collarguardrail.com/mcp-http/mcp"
    }
  }
}
Verify:

bash
curl -X POST https://api.collarguardrail.com/mcp-http/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":1}'
Available Tools
Tool	Purpose
evaluate_trade	Pre-trade risk check at fixed Tier 1 ($5,000 ceiling). Returns allow / warn / deny with reasons and audit hash. Honeypot findings (severity=danger) force a DENY.
evaluate_trade_paid	Same as evaluate_trade, gated by an x402 payment. Evaluated as Tier 2 ($25,000 ceiling). Call without payment_proof to receive the challenge; settle on-chain; retry with the proof.
check_token_safety	Honeypot / contract safety check for any ERC-20. When severity=danger, evaluate_trade auto-DENYs.
simulate_balance	Read-only balance simulation via eth_call state override
get_supported_assets	Official Robinhood Chain asset registry
verify_audit_trail	Verify the hash-chain integrity of past decisions
Call get_supported_assets first to resolve a symbol to its canonical
contract address. A mismatched address is treated as a fake-token attempt
and denied.

The free evaluate_trade is capped at Tier 1 ($5,000). If you need Tier 2
($25,000) without holding COLR, use evaluate_trade_paid — the on-chain
x402 payment is the credential.

Capabilities
Pricing

Chainlink oracle prices for 34+ Stock Tokens on Robinhood Chain

Uniswap V4 fallback when the oracle is stale, paused, or unconfigured

No fabricated fallback — if neither oracle nor V4 yields a price, verdict is deny

10-second V4 price cache

US Market Hours

Full US holiday calendar with observed-on-Friday/Monday shift

Early close on Thanksgiving Friday and Christmas Eve

Policy Enforcement

Tier system (REST API): 1,000 / 2,500 / 5,000 COLR → $5K / $25K / $100K limits

Per-tier rate limiting

Cluster-based rate limit and daily-loss guard (keyed on wallet fingerprint)

Slippage / MEV guard (hard 500 bps floor)

Idempotency via request_id (5-minute cache)

Global kill switch, contract mismatch detection, per-wallet blocklist

Contract Safety

Honeypot detection: bytecode scan, owner check, sell simulation via eth_call

Auto-DENY: when severity=danger, the verdict is deny regardless of
other checks — a hard stop.

GoPlus intelligence for unregistered tokens (buy/sell tax, mintable,
proxy, ownership reclaim).

Excessive tax (buy or sell > 10%) triggers an automatic deny via
GoPlus when allow_unregistered=true is passed to the request.

Auth & Infrastructure

JWT authentication via EIP-191 wallet signature (REST API)

Auth rate limiting per IP (20 req/min)

WalletConnect v2 + EIP-6963 wallet discovery

200+ token registry

Hash-chained audit trail (audit_hash + audit_seq)

Observability

Public live status page — uptime, dependency health, known issues

Public latency dashboard — p50/p95/p99 per endpoint and per MCP tool

Public methodology — how the 0–100 score is computed

Public audit-chain verification

Payments (x402)
Endpoint: POST /api/x402/analyze

Price: $0.10 USDG per call

Network: Robinhood Chain (eip155:4663)

Discovery: /.well-known/x402.json

Facilitator: Ultravioleta DAO

Listing: x402 Bazaar

No signup, no API key. Registered in the x402 Bazaar for automatic agent discovery.

Response Format
json
{
  "decision": "allow",
  "reasons": ["price source: oracle", "within tier 1 ceiling"],
  "tier": 1,
  "max_trade_usd": 5000,
  "calculated_notional_usd": 1452.30,
  "price_usd": 145.23,
  "price_source": "oracle",
  "risk_score": 12,
  "audit_hash": "sha256:...",
  "audit_seq": 12345,
  "daily_pnl_usd": 0.0
}
decision: allow / warn / deny — deny is a hard stop

reasons prefixed with ADVISORY: are non-blocking

price_source: oracle / uniswap_v4 / fallback_default / unavailable

error: if present, no verdict was produced — treat as a hard stop

Discovery Files
/.well-known/mcp-server-card.json — MCP tool schemas (canonical, flat)

/.well-known/mcp/server-card.json — alias (nested)

/.well-known/agent-card.json — A2A skills

/.well-known/x402.json — x402 payment resources

/.well-known/methodology.md — risk score methodology (agent-readable)

/.well-known/known-issues.json — public known-issues list

/methodology.html — methodology, human-readable

/mcp-card.html — rendered server card

/agent-card.html — rendered agent card

/latency.html — live latency dashboard

/status — live status page

/llms.txt — full API contract for LLM agents

/skill.md — concise skill index

/robots.txt — AI crawler allow-list

/sitemap.xml — site map

Registry Listings
Official MCP Registry: io.github.aiguardrail/backend

Smithery: search "collar"

x402 Bazaar: /api/x402/analyze

Documentation
Live Terminal (UI): https://collarguardrail.com

Agent Integration Guide: https://collarguardrail.com/agent-docs.html

MCP Server Card: https://api.collarguardrail.com/.well-known/mcp-server-card.json

Methodology: https://api.collarguardrail.com/methodology.html

Status: https://api.collarguardrail.com/status

Latency: https://api.collarguardrail.com/latency.html

Contact
X (Twitter): @collarguardrail

Disclosure
Collar is the risk layer for autonomous finance on Robinhood Chain.
Deterministic, auditable, and independent.

COLR is our own utility token, issued and operated by the Collar team.
It is not affiliated with, endorsed by, or issued by Robinhood Markets,
Inc. or any Robinhood entity. Robinhood Chain is a public blockchain we
build on — using it does not imply any relationship with Robinhood.