PnL Labs — Solana Trust & Risk Checks

PnL Labs — MCP server & x402 API

PnL Labs gate — no verdict, no trade. MCP: https://api.pnllabs.com/mcp Hard stops: wallet UNTRUSTED/INSUFFICIENT/UNVERIFIABLE/UNKNOWN → do not copy · token HIGH_RISK/CRITICAL/UNKNOWN → do not buy · trade HIGH_COST/UNTRADEABLE/NO_POOL/UNKNOWN → not at that size · 402/errors are not verdicts. Table: GATE.md Rule file: SKILL.md · Measured examples: docs/VERDICTS.md

Trust & risk verdicts for AI trading agents on Solana. Proof of real PnL, not peak scores. Pay-per-call via x402 (USDC on Solana) — no signup, no API key, no account.

🌐 pnllabs.com · 🔌 https://api.pnllabs.com · 🐦 @pnllabs

---

Connect in 10 seconds (hosted MCP — no install)

The server is hosted. Add it to any MCP-capable assistant with one URL:

https://api.pnllabs.com/mcp

• Claude (all plans): Settings → Connectors → Add custom connector
• Grok: grok.com/connectors → Custom
• ChatGPT / Gemini: custom MCP connector → same URL
• Also listed in the official MCP Registry (io.github.PnLlabs/pnllabs) and on Smithery.

Free tier per IP/day — when it runs out, the tools reply with structured x402 payment instructions (pay-per-call, USDC on Solana).

---

Why

Every "smart money" score you've seen ranks wallets by peak or paper gains. Peaks lie. A wallet can 100x on paper, dump into no liquidity, and walk away red — while its score still says genius.

PnL Labs recomputes what actually hit the wallet — realized SOL, in vs out — and returns a conservative, machine-readable verdict. When the numbers can't be reconstructed reliably, it says UNVERIFIABLE instead of guessing.

The four checks

Endpoint	Question	Verdicts
GET /check/wallet/{address}	Is this wallet worth copying? (real realized PnL)	TRUSTED · NEUTRAL · UNTRUSTED · INSUFFICIENT · UNVERIFIABLE
GET /check/funder/{address}	Where did this wallet's money come from? (funding origin / cluster links)	CEX_FUNDED · WALLET_FUNDED · UNKNOWN_ORIGIN · UNVERIFIABLE_ORIGIN
GET /check/token/{mint}	Is this token structurally safe?	LOW_RISK · ELEVATED · HIGH_RISK · CRITICAL · UNKNOWN
GET /check/trade/{mint}?size_sol=1	What will this trade really cost? (real slippage)	ACCEPTABLE_COST · ELEVATED_COST · HIGH_COST · UNTRADEABLE · NO_POOL

Every response carries a stable verdict enum, machine-readable reasons (code + message), confidence, flags, and schema_version. Built to be parsed by agents, not read by humans. Interactive docs: https://api.pnllabs.com/docs.

Quick example

curl https://api.pnllabs.com/check/wallet/57hECCDVNmP7jCY6XX6FxP31DQuKiwQhv2g3nxGPLPmV

{
  "schema_version": 1,
  "verdict": "UNTRUSTED",
  "confidence": "high",
  "trust": 15,
  "reasons": [
    { "code": "NEGATIVE_REALIZED_PNL", "message": "Genuinely NEGATIVE: -47.1 SOL net realized" },
    { "code": "LOW_WIN_RATE", "message": "Only 27% of coins were profitable" }
  ]
}

MCP server

This repo ships an MCP server exposing the four checks as tools (check_wallet_trust, check_wallet_forensics, check_token_safety, check_trade_cost). It calls the hosted API — no RPC keys or local setup needed.

Add to an MCP client (e.g. Claude Desktop / Claude Code):

{
  "mcpServers": {
    "pnllabs": {
      "command": "python3",
      "args": ["/path/to/pnllabs_mcp.py"]
    }
  }
}

Runs on the Python standard library — no dependencies for the free/beta tier.

Payment (x402)

Calls are priced per request and paid in USDC on Solana via the x402 protocol (HTTP 402). Your agent pays a few cents, gets an on-chain receipt — no accounts, no keys stored anywhere. Prices: wallet 0.10, funder 0.05, token/trade 0.02 USDC. Any x402-capable client (or the official x402 SDKs) can pay automatically.

Honest scope

• Wallet trust & forensics are the core edge — realized-PnL truth and funding origin, which peak-based scores and EVM-only tools don't provide on Solana.
• Token safety flags structural red flags with zero noise. Behavioral rugs (an operator selling into his own hype) are invisible to all structural checkers — for those, check the wallets involved.
• We sell risk/trust verdicts, not alpha. Never "when to buy."

License

MIT — see LICENSE.