@sint/bridge-mcp

MCP (Model Context Protocol) tool call interception and risk classification. Wraps any MCP server with SINT policy enforcement.

Install

npm install @sint/bridge-mcp

Usage

import { MCPInterceptor } from "@pshkv/bridge-mcp";
import { PolicyGateway } from "@pshkv/gate-policy-gateway";

const interceptor = new MCPInterceptor({
  gateway: myPolicyGateway,
  sessionId: "agent-session-001",
});

// Intercept an MCP tool call
const decision = await interceptor.intercept({
  tool: { name: "write_file", arguments: { path: "/output.txt", content: "data" } },
  tokenId: "tok-abc",
});

console.log(decision.action); // "allow" | "deny" | "escalate"

Middleware Mode

Drop into any MCP server as middleware:

import { createSintMiddleware } from "@pshkv/bridge-mcp";

const middleware = createSintMiddleware({
  gateway: myPolicyGateway,
  onEscalate: async (ctx) => {
    // Route to approval dashboard or human operator
    return awaitApproval(ctx);
  },
});

Features

• MCP tool call interception — transparent proxy between agent and MCP server
• Automatic tier assignment — classifies tools by risk (read vs. write vs. execute vs. shell)
• Shell detection — identifies and escalates shell execution tools
• Tool Auth Manifest (TAM) — MCP SEP-2385 reference implementation
• Tool definition signing — verify tool provenance with Ed25519
• Session tracking — forbidden combo detection across tool call sequences
• Circuit breaker — automatic lockout on abuse patterns

Part of SINT Protocol

📖 Full documentation · 🚀 Getting Started