How webhook and MCP automation works

A chart alert fires. A webhook receives it. A script calls your broker. An MCP server exposes the result to an agent. Here is how each layer connects.

Paper / sim mode throughout this guide. All examples use paper accounts. No real funds. No strategy recommendations. No returns implied.

Layer 1: The webhook

A webhook is an HTTP endpoint that accepts a POST request. Your charting platform, alert script, or any HTTP client sends a JSON payload to that URL when a condition fires.

The payload carries the action (buy, sell, close), the instrument, the quantity, and any account identifier. The receiver script on your server reads those fields and routes the command to your broker API.

Alert condition fires on your chart or script. Your platform sends a POST to your webhook URL.

Webhook receiver script on your server reads the JSON body. Parses action, symbol, qty fields.

Script calls broker API using your stored API key. Places the order under your account. No third party holds your key.

Broker confirms fill or rejects. Your script logs the response. Done.

-- example webhook payload (paper mode)
{
  "action": "buy",
  "symbol": "ETHUSDT",
  "qty": "0.1",
  "account": "paper_01"
}

No custody: the broker API key lives in your receiver script's environment variables or a secrets manager you control. The webhook URL itself is public (it just receives POSTs), but your credentials never leave your server.

Layer 2: The MCP server

An MCP (Model Context Protocol) server is a process that exposes tools to an AI agent or any MCP client. In a trading context, you write an MCP server that wraps your broker connection and publishes tools like list_positions, send_order, or cancel_order.

The agent (Claude, or any MCP client) calls those tools by name. It never sees your raw API key. The broker credentials live inside the server process. The agent gets back structured data: positions, balances, fill confirmations.

-- mcp-broker-server.py (sketch, paper mode)
from mcp_sdk import Server, tool
 
@tool(description="List all open positions")
def list_positions():
  -- your creds in env, never in the call
  return broker.get_positions()
 
@tool(description="Send a paper order (sim only)")
def send_order(side: str, symbol: str, qty: float):
  return broker.paper_order(
    side=side, symbol=symbol, qty=qty, mode="paper"
  )

From the agent's side, calling the tool looks like any function call. The agent does not know or care whether the broker is a crypto exchange or a futures broker. The MCP layer abstracts that.

Layer 3: Autoview as the webhook receiver

Autoview is a third-party tool that acts as the webhook receiver for you. You point each chart alert at your personal Autoview webhook URL (https://<WEBHOOK_URL>/hook/<your-hook-id>/). Autoview parses the payload and routes to your connected broker account using credentials you provide.

The free tier includes connecting broker demo/testnet (paper) accounts. A broker demo/testnet account connected through Autoview runs the same webhook parsing and routing as a live account. You can test the full alert-to-fill flow without any real capital at risk.

Paper / sim mode only on this site. All Trade Agent Lab demonstrations show Autoview in paper mode. We make no claims about live-mode performance, fill rates, slippage, or returns. This is a plumbing walkthrough.

Trade Agent Lab has an affiliate relationship with Autoview. If you sign up through a link on this site, we may earn a commission at no cost to you. Full disclosure is on every page.

Get Autoview free → See the paper mode walkthrough