ArcScore for agents

You are an AI agent (or you build one) and you need one answer before touching a token on Arc: is it safe to trade? That is the only thing we compute. Every number is read from the chain by our own indexer, refreshed every 5 minutes, no opinions, no paid listings, no token of our own.

1 · One call, one verdict

GET https://pre.arcscore.xyz/api/token/{address}.json  (address = token contract, lowercase 0x…)

{
  "symbol": "ARGUS", "grade": "A", "score": 85, "verdict": "ok", "tradeable": true,
  "risks": { "rug": false, "honeypot": false, "thin_market": false, "wash_pct": 12.4, "unlocked_lp": false,
             "serial_rugger": false, "lookalike": false, "deployer_linked_pct": null, "liquidity_unverified": false },
  "liquidity_usd": 663601, "holders": 19430, "holders_over_100usd": 3285, "age_days": 7.6,
  "flags": [ { "kind": "ok", "text": "transfer simulation OK" } ],
  "sources": { "liquidity": "on-chain · 2 pools", "volume": "on-chain (own index)", "change_24h": "on-chain VWAP (112/175 swaps)" },
  "confidence": 1.0, "updated": 1790953000, "url": "https://pre.arcscore.xyz/token/0xece5…html"
}

verdict: ok = grade A/B · caution = C · avoid = D/F. tradeable is a hard fact, not an opinion: false when the pool was drained/pulled, the sell simulation fails (honeypot), liquidity is under $5k, or we could not verify the liquidity on-chain. confidence (0–1) = how many of the key inputs come from our own on-chain index instead of a third party — act on 0.8+ and read the flags on anything lower. 404 = not rated (not on Arc, or under $5k liquidity).

2 · Everything at once

GET https://pre.arcscore.xyz/api/tokens.json — all rated tokens, compact, best first (address, grade, score, verdict, tradeable, liquidity, holders, honeypot, rug, wash_pct). GET https://pre.arcscore.xyz/api/index.json — endpoints and refresh info. GET https://pre.arcscore.xyz/alerts.json — last 200 events (rug, liquidity, grade changes). GET https://pre.arcscore.xyz/data.json — the full dataset the site is rendered from.

3 · Plug it into your agent

MCP (Claude, Cursor, any MCP client) — one file, two tools (check_token, list_tokens):

curl -O https://pre.arcscore.xyz/agents/arcscore_mcp.py
pip install mcp requests
# claude_desktop_config.json / .mcp.json:
{ "mcpServers": { "arcscore": { "command": "python3", "args": ["/path/to/arcscore_mcp.py"] } } }

OpenAI-style function calling — ready-made tool schema: https://pre.arcscore.xyz/agents/tool_schema.json. Plain HTTP works from anything: no key, CORS open, JSON only.

4 · Rules of the road

Be gentle: about 1 request per second and cache for 60 s (the data only changes every 5 minutes). Please credit arcscore.xyz when you show our numbers. This is descriptive on-chain data, not investment advice — your agent decides, we only tell it what the chain says. Something wrong or missing? @dexscorer.

5 · What the chain actually tells us

Liquidity depth and price impact · LP status (launchpad-locked / factory / manual) · holder count with a committed core (wallets ≥$100, ≥$1,000) and dust share · top-10 concentration and deployer-linked clusters from our own transfer index · 24 h volume, traders and round-trip (wash) share from our own swap index · sell simulation (honeypot check) · same-symbol copies · deployer history (serial ruggers) · rugs and post-launch dumps confirmed on-chain. Full method: how it works.