Agent Service Provider · A2MCP

AGT Strategy Intelligence

The reputation and risk-discipline layer for the agent economy. Honest, on-chain-verified AI trading intelligence, exposed as callable tools over MCP so any agent can score a strategy, size a trade, read the market, route yield, or backtest an idea. Every number is verifiable; no tool ever places an order.

Run the tools live OKX.AI Marketplace

Endpoint · Streamable HTTP · JSON-RPC 2.0 · MCP 2025-06-18

https://agt.money/api/mcp

Connect from any MCP client:

claude mcp add agt-strategy-intelligence https://agt.money/api/mcp -t http

Five tools

Each wraps the exact logic the AGT Money platform runs in production.

agt_scorereputation

The credit score for an AI strategy. A 0-100 AGT Score plus grade, derived only from a verified, Merkle-anchored on-chain track record. An allocator agent calls it to vet a strategy before trusting it.

swarm id → score, grade, deflated Sharpe, drawdown, value returned, epochs live

size_positionrisk desk

The risk desk other agents can rent. Turns a raw trade idea into a disciplined size with a PASS / RESIZE / VETO verdict, using volatility targeting x fractional Kelly x regime, then hard exposure and leverage caps.

confidence, leverage, vol, regime → verdict, sized position, reason

market_contextsignal

Live market intelligence from Hyperliquid: the 2d/7d trendZ regime, funding carry, basis, realized volatility, plus a directional signal that can be driven by your own edge mix.

coin (+ optional edge mix) → regime, funding, basis, vol, signal

best_poolyield router

A risk-aware DeFi yield router for Base. Ranks lending and LP pools that clear a risk tier's mandate (stablecoin-only, IL, TVL floor, reward-share cap), each scored and given a conviction.

risk tier → ranked pools with APY, TVL, score, conviction

backtestbacktest

Backtests a strategy DNA on 120 days of real hourly candles across a universe, with a buy-and-hold benchmark and a capacity estimate. Honest by design: a backtest is overfit until proven live.

edge mix, universe, tier → Sharpe, return, drawdown, vs hold

Try it live

Every button below fires a real JSON-RPC call against the endpoint above and shows the raw response. Nothing is mocked.

Reputation of a strategy, from its verified on-chain record.

Request · tools/call

{
  "name": "agt_score",
  "arguments": {
    "swarm": "demo"
  }
}

Response · live from /api/mcp

Click Run live to call the real endpoint.

Why it is different

Verified, not claimed

agt_score is gated on a Merkle-anchored on-chain track record. An unproven strategy returns 'Unproven', never a flattering fake number.

Advisory, never custodial

Every tool reads data or sizes risk. None place an order, move funds, or touch your keys. The agent never trades on your behalf.

Real logic, no drift

Each tool wraps the exact engine AGT Money runs in production, so what an agent reads here is what the platform actually does.

Cheap to call

No tool invokes an LLM. They read on-chain state, Hyperliquid, and DeFiLlama, or run pure math, so pay-per-call pricing stays honest.

Reputation, one layer down

An agent economy runs on trust between agents. AGT Strategy Intelligence applies that same idea to trading strategies: not just is this agent trustworthy, but is this specific strategy trustworthy. AGT Money is a permissionless launchpad for AI trading swarms, each proving itself with a verifiable, on-chain track record. AGT Strategy Intelligence packages that machinery as an Agent Service Provider: reputation to vet a strategy, a risk desk to stay disciplined, and live intelligence to act on. Five tools, zero LLM calls in any of them, so the surface is real infrastructure, not a chatbot wrapper. Pay-per-call settlement rides the marketplace's own stablecoin rail. Advisory only, by design.