Agent Observability

MCP Sentinel

Drop-in observability for agent tool calls. Every MCP interaction is logged, guarded, and auditable — without changing your tool implementations.

Quickstart

What

A proxy layer between AI agents and their tools. Logs every call, applies guard rails, and surfaces anomalies in a real-time dashboard.

Who

Teams deploying LLM agents in production who need visibility into tool usage, cost tracking, and security boundaries.

Why

Agents call tools autonomously. Without observability, you cannot audit decisions, detect misuse, or enforce compliance.

Quickstart

mcp-server.ts

// Reference implementation — github.com/pappdavid/mcp-sentinel
import { MCPSentinel } from './sentinel';

const sentinel = new MCPSentinel({
  apiKey: process.env.MCP_API_KEY,
  guards: ['injection', 'pii', 'cost'],
});

// Wrap your MCP server
const server = sentinel.wrap(yourMCPServer);
server.listen(3001);

Live Demo

Run a sample workflow to see how Sentinel logs and guards tool calls.

Architecture

Guard Flow

Demo is rate-limited to 100 events/minute with mock data. Contact me for a full walkthrough or to discuss deployment.