Magneta Contract Risk
Smart-contract risk reports for AI agents on 20 EVM chains, paid per call via x402 ($0.02 USDC).
Installation
Ajoutez ceci à la configuration MCP de votre client (Claude Desktop : claude_desktop_config.json ; Cursor : .cursor/mcp.json ; VS Code et les autres utilisent la même forme « mcpServers »).
claude mcp add --transport http magneta-contract-risk 'https://x402.magneta.finance/mcp'
{
"mcpServers": {
"magneta-contract-risk": {
"type": "http",
"url": "https://x402.magneta.finance/mcp"
}
}
}
Documentation
Tirée du README du projet sur GitHub, actualisée chaque semaine. Les liens ouvrent GitHub.
Sommaire
Magneta Contract Risk (x402)
A small production-oriented service that sells an on-chain smart-contract risk report to AI agents, paid per call with
x402 micropayments (USDC, EIP-3009 exact scheme). One express process, two paid surfaces:
| Surface | Route | Price |
|---|---|---|
| HTTP | GET /v1/risk/:chainId/:address | $0.02 |
| MCP (Streamable HTTP, stateless) | POST /mcp, tool contract_risk_summary(chainId, address) | $0.02 |
| MCP | tool supported_chains | free |
| Health | GET /health | free |
Both paid surfaces call the same analyzeContract(chainId, address) in src/risk/analyze.ts. The analysis is deterministic and
uses public data only (RPC, Blockscout, Sourcify, a local aderyn binary). No LLM, no API key for analysis.
Unknown chain or malformed address returns 400 before any payment is requested. If the analysis itself fails (not a contract,
upstream error, timeout) the response is an error status and the buyer is not charged (x402 only settles successful responses).
Supported chains (v1, src/chains.ts): Base 8453, Base Sepolia 84532, Ethereum 1, Arbitrum One 42161. Override RPCs with RPC_URL_<chainId>.
Report
{ chainId, address, analyzedAt, isContract, bytecodeSize, proxy, erc20, opcodes, selectors: { found },
blockscout, sourcify?, aderyn, score, level, flags, notes }
proxy:{ type, implementation, admin? }from EIP-1967 (implementation/admin/beacon), legacy zeppelinos slot, EIP-1822, EIP-1167 clones, or Blockscout's detection. The implementation bytecode is analysed too.erc20: name, symbol, decimals, totalSupply,owner()(and whether it is the zero address),paused(). Each probe is individually try/catch.opcodes: SELFDESTRUCT / DELEGATECALL presence from a real opcode walk that skips PUSH data (no substring matching).selectors.found: risky signatures (mint, pause, blacklist, fee/tax setters, upgrade, ownership...) whose 4-byte selector appears as a PUSH4 in the bytecode. Heuristic: presence is not proof of reachability (also stated innotes).blockscout/sourcify: verification status, compiler, contract name, creator and creation tx. Blockscout has a 5 s timeout; on failureblockscout: { available: false }. Sourcify is only queried as a fallback when Blockscout is unreachable or reports "unverified". For proxies, the implementation's verification is what counts.aderyn: when verified Solidity source is available, sources are written to a temp dir (sanitised paths) and aderyn runs with a 15 s timeout:{ ran, counts: { high, low }, top: [{ title, severity, instances }] }, else{ ran: false, reason }. It never fails the request.- Total budget: hard deadline 24 s per analysis. Results are cached in memory for 10 minutes per chainId+address (the payment still happens on every call).
Score rubric (src/risk/score.ts)
Score 0..100, higher = riskier, additive, capped at 100.
| Points | Condition |
|---|---|
| +25 | logic contract source not verified (Blockscout, then Sourcify) |
| +0 (flagged) | verification unknown because every provider was unreachable |
| +15 | proxy / upgradeable |
| +10 | owner() exists and is not the zero address |
| +15 | mint selector |
| +15 | blacklist / bot-list selector |
| +10 | fee / tax / limit / trading-switch setters |
| +5 (+5 if currently paused) | pause selector |
| +10 | SELFDESTRUCT opcode |
| +5 each, max +20 | aderyn high-severity findings |
Level: 0-24 low, 25-49 medium, 50-74 high, 75-100 critical. A non-contract address is an error, not a score. The score is a triage signal, not an audit.
Supported chains
Mainnets: Ethereum (1), OP Mainnet (10), BNB Smart Chain (56), Gnosis (100), Unichain (130), Polygon (137), Sonic (146),
World Chain (480), Mantle (5000), Base (8453), Arbitrum One (42161), Celo (42220), Avalanche C-Chain (43114), Linea (59144),
Blast (81457), Scroll (534352). Testnets: Base Sepolia (84532), Sepolia (11155111), OP Sepolia (11155420), Arbitrum Sepolia (421614).
Chains without a Blockscout instance (BNB, Sonic, Mantle, Avalanche) rely on Sourcify for verified sources. zkSync Era is
excluded on purpose: EraVM bytecode defeats the opcode and selector heuristics. Compiler remappings from Blockscout or
Sourcify are written to remappings.txt so aderyn can resolve @openzeppelin/...-style imports.
Payment networks
Buyers pay in USDC on Base (eip155:8453). When SOLANA_PAY_TO is set, the same routes and tools also accept USDC on
Solana mainnet (solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp), settled by the same Coinbase CDP facilitator, which
pays the Solana fees. The Solana receiving address must already own a USDC token account (the SDK does not create it).
pnpm buy-once-svm pays one call from a Solana keypair (SVM_PRIVATE_KEY, base58 64-byte secret key).
Free tier, agent skill and client package
- Free preview:
GET /v1/flags/:chainId/:addressand the MCP toolcontract_quick_flagsreturn the bytecode-only part of the report (proxy, owner/pause state, risky selectors, opcodes, partial score) with no payment. Rate-limited to 30 requests per minute per IP. Source verification and aderyn findings are only in the paid report. - Agent skill:
npx skills add Bosacks/magneta-risk-x402 --skill magneta-contract-riskinstalls skills/magneta-contract-risk/SKILL.md, which teaches an agent when and how to buy a report over HTTP, MCP or a Coinbase Agentic Wallet. - Client package:
@magnetafinance/contract-risk(npm i @magnetafinance/contract-risk @x402/core@2.28.0 @x402/evm@2.28.0 @x402/fetch@2.28.0 viem@2.57.7, source inpackages/contract-risk-client) wraps the x402 payment flow in a typed client and ships tool adapters for LangChain.js, the Vercel AI SDK and OpenAI/Anthropic function calling. - ElizaOS plugin:
packages/plugin-elizaos(@magnetafinance/plugin-contract-risk) adds the actionsCONTRACT_QUICK_FLAGS(free) andCONTRACT_RISK_REPORT(paid, capped per payment) to an Eliza agent. - ERC-8004 identity: the registration file erc8004/agent-registration.json is served at
/.well-known/agent-registration.json;scripts/erc8004-register.tsprints theregister(string)call for the Base IdentityRegistry0x8004A169FB4a3325136EB29fA0ceB6D2e539a432.
Run
pnpm install --ignore-scripts
cp .env.example .env.local # fill PAY_TO, X402_NETWORK, FACILITATOR_URL (+ CDP keys if cdp)
pnpm dev # tsx, or: pnpm build && pnpm start
pnpm type-check && pnpm test # unit tests, no network
pnpm analyze 84532 0x036CbD53842c5426634e7929541eC2318f3dCF7e # free local run, prints the report
Buying (needs a funded buyer key, spend cap $0.05; never put this key on the server):
BUYER_PRIVATE_KEY=0x... pnpm buy-once # HTTP; prints the report and the PAYMENT-RESPONSE tx hash
BUYER_PRIVATE_KEY=0x... pnpm buy-once-mcp # same through the MCP tool
Environment
PUBLIC_URL(https, no trailing slash): advertised as the MCP resource URL for Bazaar discovery. Behind a reverse proxy the server trusts the loopback proxy (trust proxy = loopback) so HTTP resource URLs are advertised ashttps://; without that, the facilitator rejects discovery with "resource must start with https://".
| Var | Meaning |
|---|---|
PORT | listening port (default 3011) |
PAY_TO | receiving address (required) |
X402_NETWORK | CAIP-2 network, default eip155:84532 (Base Sepolia); production: eip155:8453 |
FACILITATOR_URL | cdp (Coinbase CDP facilitator, needs CDP_API_KEY_ID + CDP_API_KEY_SECRET) or an HTTP facilitator URL such as https://x402.org/facilitator (testnet) |
ADERYN_BIN | aderyn 0.6.8 binary, default /home/dominique/bin/aderyn |
RPC_URL_<chainId> | optional RPC override per chain |
BUYER_PRIVATE_KEY, TARGET_URL, MCP_URL, CHAIN_ID, ADDRESS | buyer scripts only |
Deploy (systemd + Caddy)
pnpm install --ignore-scripts && pnpm build
rsync -a --exclude node_modules --exclude .env.local --exclude 'scripts/logs' ./ server:/srv/magneta/risk-x402/ # never overwrite the server's .env.local; then pnpm install --prod --ignore-scripts there
sudo cp deploy/magneta-risk-x402.service /etc/systemd/system/ && sudo systemctl enable --now magneta-risk-x402
# add deploy/Caddyfile.snippet to the Caddyfile (vhost x402.magneta.finance -> 127.0.0.1:3011), reload Caddy
The unit reads /srv/magneta/risk-x402/.env.local. For mainnet set X402_NETWORK=eip155:8453 and FACILITATOR_URL=cdp.
Aderyn needs its solc download/cache available to the service user (HOME writable).
Bazaar discovery
Both paid surfaces declare x402 Bazaar discovery metadata with declareDiscoveryExtension from @x402/extensions/bazaar
(HTTP route with path-parameter schema and output example; MCP tool with input schema), and the bazaarResourceServerExtension
is registered on the resource server. The metadata appears in the PAYMENT-REQUIRED header (extensions.bazaar). Actual
cataloguing depends on the facilitator: it indexes resources when it settles a payment, so the listing appears only after the
first real paid call through a Bazaar-enabled facilitator (CDP). The public x402.org testnet facilitator is not expected to list it.
Production status (2026-10-16)
Live at https://x402.magneta.finance on Base mainnet (eip155:8453), Coinbase CDP facilitator, $0.02 per call, listed in the
x402 Bazaar (route /v1/risk/:chainId/:address). Buyers need USDC on Base and no ETH. The facilitator rejects payments where
the payer equals payTo (self_send_not_allowed).
License
MIT, Copyright (c) 2026 Magneta Finance. See LICENSE.
Questions fréquentes
Qu'est-ce que Magneta Contract Risk ?
Smart-contract risk reports for AI agents on 20 EVM chains, paid per call via x402 ($0.02 USDC).
Comment installer Magneta Contract Risk ?
Il fonctionne comme serveur distant. Dans Claude Code : claude mcp add --transport http magneta-contract-risk 'https://x402.magneta.finance/mcp'. Pour les autres clients, collez la configuration JSON de cette page dans leurs réglages MCP.
Faut-il une clé API pour Magneta Contract Risk ?
Son entrée au registre n'indique aucune clé ni secret obligatoire. Certaines fonctions peuvent tout de même demander un compte auprès du service auquel il se connecte.
Magneta Contract Risk est-il open source ?
Oui. Son code est sur GitHub sous licence MIT.
Qui maintient Magneta Contract Risk ?
Il est publié par Bosacks. AI.tn le référence sans le maintenir ni le recommander.
Issu du registre MCP officiel. Être listé n'est pas une recommandation : vérifiez le code et les permissions d'un serveur avant de lui donner accès à vos données.