Stop your AI agent
from getting rugged
AI agents trading on-chain can't tell safe contracts from scams. Aegis checks every contract, simulates every trade, and blocks dangerous transactions before your agent loses funds.
npx aegis-defi
Click to copy
Cumulative Agent Losses
$62,400,000
warning AI agent exploits increased by 420% in Q4 2025
The price of
unsecured intelligence.
Current agent architectures lack sovereign safety layers. Aegis intercepts malicious logic before it hits the mempool, preventing catastrophic capital drainage.
-
shield
Pre-flight Simulation
Fork the chain. Run the trade. Check the outcome. All before real funds move.
-
layers
Multi-Layer Verification
Static analysis, bytecode scanning, fork simulation, on-chain enforcement.
How It Works
Three layers of protection
MCP Safety Server
One command adds six safety tools to Claude, GPT, or any MCP-compatible agent. Your agent calls scan_contract, simulate_transaction, check_token, assess_risk, trace_transaction, or search_solodit before every trade.
Contract Scanner
165 exploit pattern checks against source code and bytecode.
Fork Simulation
Run the trade on a forked chain. Check gas, slippage, revert risk. No real funds at risk.
On-Chain Enforcement
Even if software fails, the blockchain doesn't. AegisGateway verifies a signed safety attestation before executing any DeFi call. No attestation, no execution.
Learn more arrow_forward# Add Aegis to Claude Code
$ claude mcp add aegis npx aegis-defi
# Your agent now has:
# scan_contract
# simulate_transaction
# check_token
# assess_risk
# trace_transaction
# search_solodit
The Safety Pipeline
Every trade passes through Aegis
Agent Intent
Contract Scan
Simulation
Execution
What It Catches
Known scams, caught instantly
Before your agent touches a contract, Aegis checks it against known attack patterns in both source code and bytecode.
+ 10 more patterns including delegatecall injection, oracle manipulation, gas griefing, and cross-chain replay. See all
Integration
Add it in 30 seconds
Aegis is an MCP server. If your agent speaks MCP, it already works.
# Add Aegis to Claude Code $ claude mcp add aegis npx aegis-defi # Or run directly $ npx aegis-defi # Your agent now has 6 safety tools: # scan_contract - Static analysis for 165 exploit patterns # simulate_transaction - Forked-chain tx simulation # check_token - Anti-honeypot token checks # assess_risk - All-in-one risk assessment # trace_transaction - Multi-contract call tree analysis # search_solodit - Cross-ref 50K+ real audit findings
{ "mcpServers": { "aegis": { "command": "npx", "args": ["aegis-defi"] } } }
import { createAegisServer } from "aegis-defi"; // Create and start the MCP server const server = createAegisServer(); // Connect via stdio transport const transport = new StdioServerTransport(); await server.connect(transport); // The server exposes 6 tools: // scan_contract, simulate_transaction, // check_token, assess_risk, // trace_transaction, search_solodit // assess_risk returns signed attestations for // ALLOW/WARN decisions (chainId-bound signatures)
On-Chain Enforcement
Smart contracts that can't be skipped
Off-chain checks can be bypassed. These can't. Two smart contracts on Base enforce safety at the protocol level.
AegisGateway Base
For agents. Instead of calling DeFi contracts directly, your agent routes calls through the Gateway. It checks for a valid safety attestation (signed by Aegis after scanning and simulating) before forwarding the call. No attestation = transaction reverts.
AegisSafetyHook Uniswap v4
For pool creators and LPs. Install this hook on your Uniswap v4 pool to require Aegis-verified attestations before any swap. A scam pool would never install this -- its presence is itself a trust signal.
Real-World Impact
$62M lost. All preventable.
Real incidents from the last year. Each one would have been blocked by Aegis before any funds moved.
Aave Swap Disaster
March 2026 -- A whale swapped $50M USDT for AAVE tokens. Routed through a pool with only $73K liquidity. MEV bots extracted $34M.
Cork Protocol Hack
May 2025 -- Missing access controls in CorkHook.beforeSwap() on Uniswap v4. Attacker bypassed all safety checks.
Moonwell Oracle Attack
Feb 2026 -- Oracle misconfiguration on Base priced cbETH incorrectly. Attacker exploited the price discrepancy.
Architecture
End-to-end safety pipeline
GPT
Any agent
scan_contract
simulate_transaction
check_token
assess_risk
trace_transaction
search_solodit
Static analysis
Bytecode check
Fork simulation
attestation
(chainId-bound)
attestation
on-chain
swap
Uniswap v4
Ready to secure
your agent?
Open source, MIT licensed, free to use. One command to protect every trade your agent makes.