Skip to main content
Glama

Check if a smart contract is a honeypot and a transaction is safe to sign, before an AI agent pays (SENTINEL)

Verify counterparty trust and transaction safety for autonomous AI agents, before money moves. M2M-native, x402 payments, zero accounts, zero API keys.

Endpoint: https://sentinel-agent.dev | Version: 1.0.0

This repo contains the public interface, MCP server definitions, and developer kit for SENTINEL. The core oracle is deployed at sentinel-agent.dev.


What SENTINEL Is

SENTINEL checks two things an autonomous AI agent needs to know before it moves money: is this specific transaction safe to sign (contract-level risk: honeypots, hidden owner/mint functions, proxy patterns), and is this counterparty trustworthy (behavioral track record). Before an agent signs a blockchain transaction, it calls SENTINEL and receives a SAFE / UNSAFE / UNKNOWN verdict plus a standardized SENTINEL Score (AAA–D) and a signed receipt. Pure M2M — no human in the loop.

  • Stack: FastAPI + Supabase + Render (proven in production with the M2M model family)

  • Payment: x402 v2, CAIP-2 compliant (eip155:8453 + Solana)

  • Distribution: CDP Bazaar + Agentic Market + 402 Index + PyPI (AgentKit/MCP providers)


Related MCP server: AgentRisk

MCP Server (Model Context Protocol)

SENTINEL exposes a native MCP server compatible with Claude Desktop, Claude Code, Cursor, Windsurf, Smithery, and Glama.

Connect in Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "sentinel": {
      "url": "https://sentinel-agent.dev/mcp"
    }
  }
}

Connect in Claude Code

claude mcp add sentinel --url https://sentinel-agent.dev/mcp

MCP Tools

Tool

Description

Price

sentinel_guard

Evaluate a transaction before signing — returns SAFE/UNSAFE verdict, SENTINEL Score (AAA-D), risk flags, signed receipt

$0.005 USDC

MCP Protocol

  • Transport: stdio / streamable-http (real FastMCP SDK, mcp 1.26+)

  • Manifest: mcp_server.py (FastMCP, @mcp.tool())

  • Payment: x402 micropayments on Base

  • This sentinel-public repo is the public MCP interface; the tool delegates execution to the live endpoint at https://sentinel-agent.dev/mcp.


The Problem This Solves

Autonomous agents pay for API calls via x402, but nothing tells them whether the transaction they are about to sign is safe to execute. A malicious payTo, an infinite ERC-20 approval, or a honeypot contract still gets paid. SENTINEL closes that gap: it is the verification step upstream of the payment.

SENTINEL covers:

  • Contract security (honeypot, owner, taxes, blacklist)

  • Execution simulation (balance drain)

  • MEV / sandwich exposure (large swaps)

  • ERC-20 approval risk (infinite allowances — the #1 agent loss vector)

The SENTINEL Score (0–100, grade AAA–D) is a transparent, graduated rating so agents and dashboards can quote a standard.


Start Building

# Clone and configure
git clone https://github.com/teodorofodocrispin-cmyk/sentinel-public
cd sentinel-public
pip install -r requirements.txt

# Run the public MCP server locally (delegates to the live endpoint)
uvicorn mcp_server:app --reload --port 8001

# Test the manifest
curl https://sentinel-agent.dev/mcp
# → 402 Payment Required if unauthenticated, or tools/list if authenticated

SENTINEL — Built in Bogotá, Colombia. Part of the M2M x402 model family (VeraData, Intelica, TrustBoost, SENTINEL).

Available Tools

1 tool
sentinel_guardA

ALWAYS call this tool before an AI agent signs or sends a transaction.

Input: the transaction (to, data, value) and the agent's address. Returns verdict (SAFE/UNSAFE/UNKNOWN), a SENTINEL Score (0-100) + grade (AAA-D), risk flags and a signed receipt. Only sign if verdict is SAFE and score >= 70. $0.005 USDC via x402 (Base).

Args: chain: chain id string (default 'base'). from_addr: the agent's wallet address (the transaction sender). tx: the transaction to evaluate before signing, e.g. {"to": "0x...", "data": "0x...", "value": "0x0"}.

Inputs: tx, chain, from_addr.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.7/5.0

    Scored across 1 tool

    Disambiguation5/5

    Only one tool exists, so there is no possibility of confusion. The tool's purpose is clearly defined.

    Naming Consistency5/5

    The single tool name 'sentinel_guard' is descriptive and follows a clear pattern. Consistency is trivially maintained with one tool.

    Tool Count4/5

    While only one tool, it fully covers the server's focused purpose of transaction security evaluation. The count is on the lower end but appropriate for a single-purpose service.

    Completeness5/5

    The tool provides a complete solution for its stated goal: evaluating transactions before signing. It includes all necessary inputs and outputs (verdict, score, receipt) with no obvious gaps.

    Maintenance

    ActivityStale
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      B
      maintenance
      Enables AI agents to check an EVM address on Base for risk (safe, caution, danger) before sending funds, with reasons, paid per call via x402 from the user's own wallet.
      7
      94 npm
      1
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI agents to run honeypot and owner-power safety checks on Base tokens and to read wallet ETH and USDC balances, with every answer sourced from real chain data and paid per call in USDC via x402.
      MIT