The launch-lifecycle risk desk for Base — with a verifiable hit rate.

Any agent can run a honeypot check. RugSense does what a freshly-prompted agent can't: catches the ~35% of rugs that have no code-level risk via wallet-behavior signals, follows every call to its outcome for a leakage-free track record, and pushes you a webhook when a position turns. One x402 call. No keys, no signup.

Catches code-clean rugs

~35% of rugs have no code-level risk — pure unlocked-liquidity pulls a honeypot check misses entirely. RugSense reads wallet behavior: funding clusters, sniper bundles, graph centrality, deployer history.

A hit rate you can audit

Every verdict is snapshotted at score time and graded strictly later — no hindsight inflation. The precision numbers are public at /api/track-record and /api/history.

Push, not pull + a moat that compounds

Register a token and we webhook you the moment it turns. And a proprietary repeat-offender denylist, grown from rugs we catch — the next launch by a known rugger is flagged before it pulls.

Live radar — latest launches, quick view. Full onchain safety + deployer/reputation ships with the API.

HOT

WATCH

AVOID

No launches resolved right now — try again shortly.

Click any token for its full report. The teaser uses fast scoring; the API runs the deep onchain assessment.

Add to your agent in one line

Drop the MCP server into Claude Desktop, Cursor, AgentKit or LangChain. Your agent's wallet pays per call over x402 — no API keys, no signup.

{
  "mcpServers": {
    "rugsense": {
      "command": "npx",
      "args": ["-y", "rugsense-mcp"],
      "env": { "BUYER_PRIVATE_KEY": "0x<a Base wallet with USDC>" }
    }
  }
}

Seven tools — get_base_launches, check_base_token, quick_check_base_token, check_base_tokens_batch, watch_base_token, get_base_deployer_dossier, get_rugsense_track_record. Or call the HTTP API directly:

GET /api/quick/{address}                         → fast pre-screen one token  $0.005
GET /api/token/{address}                         → deep-score one token       $0.03
GET /api/tokens/batch?addresses=0x..,0x..        → pre-screen up to 20         $0.10
GET /api/launches/latest?tier=HOT&minSafety=60   → ranked scored feed          $0.03
GET /api/watch/{address}?callback=https://..     → webhook on tier change/rug  $0.05
GET /api/deployer/{address}                      → accumulated deployer dossier $0.02
GET /api/track-record   ·   GET /api/history     → verifiable hit rate          free

How an agent gates on it:

const { token } = await (await pay(`/api/token/${addr}`)).json();
if (token.tier === "AVOID" || token.composite < 60) return skip();
swap();  // only HOT/WATCH with enough confidence

Discoverable on the x402 Bazaar / Agentic.Market · OpenAPI · npm · source

How the score works

Deterministic, no LLM. Every launch runs through 14+ behavioral & contract signals; a single dangerous one (honeypot, pullable LP, serial deployer, repeat offender) caps the composite to AVOID no matter the momentum.

Honeypot + tax

buy/sell trade simulation

Deployer reputation

serial-deployer + prior-token outcomes

Repeat-offender denylist

wallets tied to rugs we caught

Sniper / bundle

supply grabbed in the opening blocks

Funding cluster

top holders funded from one source

Graph centrality

one wallet seeding many holders

Latent honeypot

owner-mutable sell/tax switches

Source verified

contract is open-source

Proxy / upgradeable

can the logic change?

Mint / blacklist / pause

dangerous functions in bytecode

Ownership

renounced, or owner can act?

Holder concentration

top-10 % + whale count

LP burn / lock + duration

pullable? permanent or timed? (v2 + v3)

Momentum

liquidity, volume rotation, buyers

Every launch returns the per-signal checks[] (pass/warn/fail/unknown) + a safetyConfidence — see why, not just a number.

Receipts, not promises →

A public, timestamped record of launches we flagged AVOID that later rugged. Watch it fill.

/caught

Scores are a risk filter, not a guarantee — DYOR. Not financial advice. · open source