agentsafe

EVM smart-contract AI risk screening MCP server — Ethereum Mainnet, payable per call with x402 micro-payments in USDC on Base.

GitHub
1

Documentation

AgentSafe — AI-Powered Smart Contract Security Intelligence

x402 Pay-per-call Network: Ethereum Mainnet MCP Server

Secure your Web3 journey with intelligent vulnerability detection, multi-engine security analysis, and actionable risk insights. AgentSafe combines advanced AI reasoning with industry-leading security tools to help developers and AI agents identify potential vulnerabilities in Ethereum smart contracts.

Seamlessly integrated with MCP and powered by x402 pay-per-use USDC micropayments on Base.

🔒 Security Notice: AgentSafe provides AI-assisted security assessments. Results are informational, require independent verification, and do not replace a professional security audit.

Highlights

  • 🔍 Closure-source / deep-logic analysis — send any 0x contract address
  • ⚡ x402 pay-per-call — pay in USDC on Base, no subscription, no lock-in
  • 🧵 Streamable HTTP MCP transport — stateless & serverless-friendly
  • 🛡️ Fail-closed — payment verified on-chain before any audit is dispatched
  • ☁️ Serverless persistent — task state survives cold starts (Upstash Redis)

Validation Engine

Every candidate finding passes a deterministic validation pipeline before it reaches the report:

  • 5-state machine — confirmed (independent dual-source corroboration only), potential (default exit), informational (standard behavior / theoretical), rejected (deterministically refuted, e.g. Checks-Effects-Interactions verified in source), needs_manual_review (evidence untraceable — hallucination guard). Entry states (tool_detected/llm_suggested) never appear in final reports.
  • Anti-hallucination gate — LLM findings must carry an evidence quote that is located verbatim in the analyzed source, with line numbers; otherwise they are downgraded for manual review, never presented as fact.
  • Evidence-based severity rubric — severity is assessed from exploit prerequisites and mitigating controls, never from LLM confidence or repetition: unvalidated claims are capped at medium (the original claim is preserved in proposed_severity), confirmed findings may only be downgraded (privilege gates, pausable/timelock/guard controls), never upgraded.
  • Full audit trail — every state change records the rule that fired (re0_cei_satisfied, ov0_theoretical_only, cr0_independent_support, …), the reason, and supporting/contradicting code references with line numbers.

AgentSafe delivers AI-assisted security assessment with deterministic falsification — informational by design, not a substitute for a formal audit.

Pay-per-call Pricing

TierPriceScope
MCP audit$0.50 / $2.50 / $5.00 / $25.00by contract size
REST audit (POST /api/audit)$3.00 flat≤ 500 source lines

Connect (MCP client)

  • Endpoint: https://x402.workfern.com/mcp
  • Transport: Streamable HTTP (JSON responses)
  • initialize is free; tool calls trigger an x402 payment in USDC on Base.

REST API

curl -X POST https://x402.workfern.com/api/audit \
  -H "content-type: application/json" \
  -d '{"address":"0xdAC17F958D2ee523a2206206994597C13D831ec7"}'
# -> 402 + PAYMENT-REQUIRED header (x402 v2) with exact on-chain payment terms

OpenAPI spec: https://x402.workfern.com/openapi.json

Topics

#SmartContractAudit #Web3Security #MCP #Ethereum #x402 #AISecurity #Solidity

Refund policy

https://x402.workfern.com/refund