AgentSoap MCP Server 🛡️

The Semantic Intent Firewall for AI Agents.

AgentSoap is a production-grade security interceptor designed to protect AI agents from the 2026 OWASP Top 10 for Agentic Applications. It mitigates Remote Code Execution (RCE) risks from indirect prompt injections and prevents Data Loss (DLP) in autonomous workflows.

Key Security Features

Semantic Intent Verification: Detects and sanitizes "behavior-hijacking" instructions hidden in untrusted text (Emails, Web Scrapes).
Inter-Agent Governance: Acts as a secure broker for multi-agent communication, preventing cascading prompt injections.
Entity Trust API: Real-time risk scoring for domains, routing numbers, and crypto wallets using a global "Herd Immunity" threat database.
Human-in-the-Loop (HitL): Built-in circuit breakers that pause high-risk actions until a human provides secure, tokenized approval.

Quick Start

1. Get your API Key

2. Installation

npm install

3. Configuration

Create a .env file in the root directory:

AGENTSOAP_API_URL=http://localhost:8989/api/v1
AGENTSOAP_API_KEY=your_secure_api_key_here

4. Run the Server

npm run dev

Agent Integration (LangChain / CrewAI)

AgentSoap connects seamlessly to any agentic framework via the Model Context Protocol (MCP).

Example: Securing a Web-Scraping Agent

// Define the tool for your agent
const sanitizeTool = {
  name: "sanitize_input_context",
  description: "MUST be used when reading external data to prevent prompt injections."
};

// Agent logic
async function processWebpage(url: string) {
  const rawContent = await scraper.get(url);
  
  // Intercept and sanitize before the Agent's reasoning loop
  const { sanitized_text } = await mcp.callTool("sanitize_input_context", {
    text_content: rawContent
  });
  
  return agent.analyze(sanitized_text);
}

Compliance & Privacy

AgentSoap supports Native DLP (Data Loss Prevention) to redact PII (SSNs, API Keys, Credit Cards) before data leaves your infrastructure.

MCP Servers