v0.1.0

Quickstart Guide

Learn how to route your agent's LLM tool calls through the Out-of-Process MCP Gateway for strict Human-In-The-Loop approval.

1. The Problem

Giving an autonomous agent direct access to a mutable API (like Stripe or AWS) is dangerous. If the agent hallucinates, it executes irreversible actions. The Saga Gateway acts as a chokepoint, intercepting JSON-RPC calls before execution.

2. Wrapping your Agent

Simply point your agent's MCP base URL to the Gateway. No SDK rewrite required. The integration is seamless.

agent.ts
const client = new MCPClient({
  baseUrl: "https://api.sagagateway.com/v1/mcp",  headers: {
    "Authorization": "Bearer your_access_token"  }
});

3. The Security Lifecycle

1. Intelligent Interception

The agent's tool call is securely trapped at the edge before it can hit the target platform.

2. Pre-Execution Sandbox

Our engine evaluates the intended action, checking against API edge cases, rate limits, and structural integrity.

3. Human-In-The-Loop Verification

You review a simple, human-readable summary of what the agent is trying to do, and securely approve or deny the action.

4. Secure Dispatch & Memory

Upon approval, the Gateway executes the action via the Saga Pattern and logs it against our proprietary Reversal Matrix. If your agent hallucinates later, you have a guaranteed compensating transaction (a 'Cmd+Z' rollback) ready to fire.

4. Coming Soon: Consumer Wallets

Currently, Saga Gateway focuses on B2B infrastructure protection (e.g., stopping an autonomous support agent from accidentally refunding your customers via Stripe).

However, Consumer Wallets & Browser Extensibility are launching soon. This will allow everyday people to safely give their personal AI agents the ability to purchase items, book flights, or pay for services online using a virtual wallet, protected by the exact same Human-In-The-Loop approval process.