New research: Price-oracle manipulation: what a pre-deployment check can catch, and what it can’t
Report an issueGet a demo

Verify Every Transaction Before It Reaches the Chain

MagiChain sits between the agent, the signature and the chain. It checks every AI-agent payment against what the user actually signed for, and we are bringing the same pre-deployment check to smart contracts. Policy alone doesn’t make Web3 safe. Intent does.

This opens your email app. Or write to [email protected]. We only use your email to reply (privacy).

See how it works
010110

Built on the open standards agent payments and smart contracts run on

  • x402
  • EIP-3009
  • EIP-712
  • AP2
  • MCP
  • Base
  • EVM
  • Solidity
The threat

Agents now hold wallets. Contracts hold billions. Attackers know it.

Intent, not just policy.

Allowlists and spending caps stop the obvious theft. The dangerous transactions pass every policy: the right shop, under the limit, valid code, and still not what the user asked for.

Injection

Poisoned content

Web pages, tool descriptions and MCP replies that quietly rewrite who the agent pays and how much.

Agent Payment Verification
Merchant

Lying payment terms

402 responses with a swapped payee, a look-alike token, an inflated price or a validity window built for replay.

Agent Payment Verification
Policy-compliant

Harm inside the rules

Overpaying under the cap, extra purchases that each look fine, and payments split to slip under review.

Agent Payment Verification
Contract

Malicious contracts

Reentrancy, unsafe delegatecall, tx.origin checks and drains hidden in code that looks harmless before it is deployed.

Smart Contract Verification · in design
The platform

One verification layer, two places it matters

Prototype

Agent Payment Verification

Before an AI agent’s x402 payment is signed, MagiChain checks the merchant’s terms, the spend rules and, above all, whether the order matches the mandate the user actually signed (AP2).

  • Real AP2 intent mandates, verified on the exact terms about to be signed
  • Price, payee, token and validity checks on every 402
  • An extra layer in front of the wallet or signer you already use
Learn more about Agent Payment Verification
+Verification locks
  • C1Tool allowlist
  • C3Cashier rules
  • C4Human approval
  • C5x402 spend controls
  • C6Price & terms check
  • C7Wallet policy
  • C9AP2 intent mandateIntent
In design

Smart Contract Verification

An antivirus-grade check for smart contracts before they reach the chain. Upload a contract, get a report on what it can really do, and deploy only the version that passed.

  • Rule-based analysis of the compiled bytecode for known dangerous patterns
  • AI review, malware signature matching and wallet reputation
  • A report with warnings, violations, a score and a recommendation
Learn more about Smart Contract Verification
+Planned pipelineroadmap
  1. UploadContract source
  2. ValidateBytecode and ABI
  3. ReportVerified bytecode
  4. DeployTo the chain

Planned checks

ReentrancyUnsafe delegatecallEther leaksOwner checkstx.origin authStorage invariants

How MagiChain helps: from request to settlement

One verification layer between the agent, the signature and the chain.

  1. 01

    Intercept

    Catch the request

    The agent receives an x402 “payment required”, or a team submits a contract to deploy.

  2. 02

    Check the shape

    Run the policy locks

    Approved shops, budget and caps, price and terms, x402 spend controls and the wallet’s own policy.

  3. 03

    Check the intent

    Compare with the mandate

    The exact order is compared with what the user actually signed for, using a real AP2 intent mandate.

  4. 04

    Sign or stop

    Return a verdict

    Only what passes every lock reaches the signer. Every verdict is logged with its reason.

See every lock and what it checks

Built for

Security for the teams moving money on-chain

Agent platforms

Give your agents spending power without giving attackers a steering wheel.

Wallets

Screen every agent-initiated signature request before your user’s funds move.

Payment rails

Put an intent check in front of x402 settlement and make machine payments safe by default.

Protocol & dApp teams

Verify contracts before they reach the chain, and deploy only the version that passed.

Roadmap

From agent payments to the protocol

We start where the risk is newest, AI agents that pay on their own, and move down the stack until verification is part of the chain itself.

  1. NowPrototype

    Agent payment verification

    Intent, terms and spend checks in front of the signer. Working prototype, now onboarding design partners.

  2. Stage 1In design

    Verification gateway and API

    A web app, CLI and API: connect a wallet, upload a contract, get a security report, then deploy.

  3. Stage 2Roadmap

    Layer-2 validity proofs

    Run validation off-chain and attach a STARK-style proof that it ran, verified when the contract reaches the chain.

  4. Stage 3Roadmap

    Protocol-level validation

    Validators take part in checking contract contents, rewarded for every verification they perform.

Who we are

MagiChain Security is an early-stage security startup in Tel Aviv, building a verification layer for Web3 in two phases. We started with research into one question: can an AI agent be talked into a payment its user never wanted? Next we take the same idea to the blockchain itself: nothing reaches the chain unchecked.

Phase 1Prototype

AI-agent payments

Agents now pay on their own. We check every payment against what the user actually signed for, before it is signed.

Learn more about AI-agent payments
Phase 2In design

The blockchain layer

Verifying smart contracts before they are deployed, then moving that check into layer-2 proofs and the protocol itself.

Learn more about The blockchain layer

Investors, design partners and researchers: [email protected]

Ready to see MagiChain in action?

We’re onboarding a small group of design partners building agentic payments and on-chain products.

This opens your email app. Or write to [email protected]. We only use your email to reply (privacy).