Can an AI agent actually place my TradingView trades?

Autoview · Guide · Updated October 6, 2026

If you've searched "TradingView MCP" and landed here looking for a straight answer, here it is. An AI agent still doesn't read your TradingView alerts automatically -- nothing in Autoview's agent surface parses a TradingView payload. But since TradingView shipped its own MCP server, an agent that holds both connectors at once can now close the loop for you: it can place a real order directly through Autoview, right now, on Kraken and Crypto.com (or a free paper fill on five other venues), or it can build a TradingView alert on Autoview's side and hand it straight to TradingView's create_alert -- the two paths meet, even though neither one reads the other's data.

Two paths that now meet

Autoview runs two separate mechanisms that both end in a real order, and neither parses the other's data:

• The webhook path, the one Autoview has run for years: a TradingView alert fires, POSTs to your Autoview webhook URL, and Autoview places the order. No AI agent reads or touches any part of it. If this is what you're picturing, the TradingView webhook guide is the one you want, and it's been live the whole time.
• The MCP path, new and in early access: an AI agent you're talking to, Claude Desktop, a claude.ai connector, Claude Code, or a custom client, calls Autoview's entry tool with a structured order (exchange, symbol, side, quantity, and so on). Nothing about the tool cares whether the idea behind that order came from a TradingView chart you showed the agent, your own analysis, or a strategy you described in plain language. The agent decides what to send; Autoview validates it, checks what you're entitled to trade, and places it.

What's changed is TradingView now runs its own MCP server too, with tools including create_alert. An agent that's connected to both Autoview's MCP and TradingView's MCP in the same conversation can use Autoview's alert tool to build a webhook, message, and expiration, and pass that straight to TradingView's create_alert -- wiring up a live TradingView alert without you opening TradingView at all. Neither server reads the other's payload; the agent is the thing carrying data across the two calls. See building the round trip below for what that actually looks like.

So "can an AI agent place my TradingView trades" is really two questions stacked together. If it means "will Autoview auto-execute a TradingView alert that already exists," the answer is no, and it doesn't need to be yes, because the webhook path already does that without an agent. If it means "can an AI agent I'm using place a real trade through Autoview, or set up a TradingView alert for me," the answer is yes to both, today, and the rest of this page covers exactly what that looks like.

Building the round trip

This needs an agent holding two separate connections: Autoview's MCP and TradingView's MCP (mcp.tradingview.com/mcp, TradingView's own OAuth-protected server). Ask it to turn an order into a standing TradingView alert, and here's what it does under the hood:

1. Calls Autoview's alert tool with your hook_id and order details. Autoview hands back message, webhook, expiration, and resolution -- everything create_alert needs except the symbol and price condition.
2. Converts the symbol. Autoview's alert returns your symbol exchange-native (XBTUSD on Kraken), but TradingView's create_alert wants its own EXCHANGE:TICKER form (KRAKEN:XBTUSD). The agent has to make this conversion itself -- neither tool does it for the other.
3. Calls TradingView's create_alert with the converted symbol, a price condition (TradingView's alert conditions are price-only, not the arbitrary logic a Pine Script strategy can express), and the webhook, message, expiration, and resolution Autoview supplied.

A handful of TradingView-side facts matter here, none of them under Autoview's control: TradingView's MCP server requires an Essential plan or above (trial plans are excluded); create_alert's own expiration defaults to 30 days out if the agent doesn't pass one (Autoview's alert defaults to a year, precisely so an agent forwarding it doesn't fall back to TradingView's shorter default); the webhook argument requires a TradingView plan with webhooks enabled plus two-factor authentication on the account; get_alerts strips message and webhook from what it returns, so checking a live alert's actual payload later means calling get_alerts_log instead; update_alert can swap the webhook URL but cannot change the condition, symbol, or timeframe, so a real edit means deleting and recreating the alert; and TradingView's MCP docs put the rate limit at roughly 100 tool calls per minute per user.

Two more limits come from TradingView's own beta announcement. It says market data from the MCP server is delayed, and that TradingView may limit the number of requests per day while the beta runs. It doesn't give a figure for either one. The delay matters most when an agent reads a price on TradingView and then calls Autoview's entry with a limit price worked out from it, because that price can already be behind the market. Before you let an agent trade live on a number it got from TradingView's MCP, check that number against your exchange.

Once it's created, the alert runs exactly like any other TradingView alert: it fires on TradingView's servers, POSTs to the webhook URL Autoview gave it, and Autoview's existing webhook path takes it from there, with no agent involved again unless you ask it to check or edit the alert later.

What about TradingView's own AI Copilot?

In late September 2026 TradingView announced AI Copilot as a native part of Supercharts. It's a separate paid add-on for Essential and every plan above it, and it sits in the side panel next to your chart. TradingView says it can create alerts and write and apply Pine Script for you. The announcement doesn't say it can place orders, and it doesn't mention brokers or webhooks.

That makes Copilot a way to set up an alert, not a way to execute one. It's also a different thing from the round trip above: there, an outside agent calls two MCP servers, while Copilot lives inside TradingView's own side panel. Autoview never talks to Copilot. It only sees the webhook an alert sends when it fires, so if an alert Copilot made carries your Autoview webhook URL and a valid Autoview command in its message, Autoview handles it exactly like an alert you built by hand.

Don't take that part on trust. After Copilot creates an alert, open it for editing in TradingView and confirm your Autoview webhook URL is set and the message holds the command you expect. If the URL isn't there, that alert won't reach Autoview. Webhook alerts on TradingView need Essential or above plus two-factor authentication, whoever wrote the alert. TradingView is also only one place an alert can come from. Any tool that can send a webhook can drive Autoview, and the alert automation guide covers the webhook setup itself.

Live venues and paper venues

Two venues place real money orders through Autoview's MCP server today:

• Kraken -- live, requires an active Autoview Kraken subscription (the same one that gates Kraken on the webhook path).
• Crypto.com -- live, free, under Autoview's partnership lane.

Four more are free paper accounts you can connect, useful for watching an agent work before it ever touches funded money:

• Tradovate Simulation, OANDA Practice, Bybit Testnet, and Bybit Spot Testnet -- all $0, all real fills against play money.

For live trading on OANDA or Bybit, use the webhook path. Through the MCP, those two connect to their paper accounts.

Why a misread instruction previews instead of trading

Every call to entry is a dry run by default. An agent can misunderstand an instruction, so nothing gets sent to a broker unless you explicitly set live to true on that specific call. A dry run still builds the exact order Autoview would have placed and hands it back for you to read, so you can check the symbol, side, size, and any stop before anything real happens. That default applies the same way on Kraken and Crypto.com as it does on a free paper account. Going live is always a deliberate, separate step, never the fallback. But be clear about who takes that step: live is a field on the call, and the agent is the one making the call. With a dry-run-only token, Autoview refuses any live call. With a live-enabled token, the agent can set live itself, so a misread instruction on that token can become a real order.

Tokens add a second layer: each one is scoped at creation to specific exchanges and to dry-run-only or live-enabled, and you can revoke any token instantly from /account/tokens/, which cuts an agent's access immediately, mid-session if needed.

Which AI agents actually connect

Anything that speaks the Model Context Protocol. Claude Desktop, the claude.ai connectors, and Claude Code all handle the connection handshake for you out of the box. A custom or autonomous agent client works too, as long as it completes the standard initialize call and carries the Mcp-Session-Id header the server hands back on every request after that, the one piece the ready-made clients hide from you.

Matching each job to the right path

For a protective stop and target on Tradovate Simulation, attach stopLoss and takeProfit, or a trailing stop, to a market entry. On Kraken and Crypto.com, place the protective stop as its own order, with type stop and a trigger price. For an alert that already exists in TradingView, put your Autoview webhook URL and command on it, and the webhook path runs it with no agent involved. The round trip above is for building new alerts. For entry logic beyond a price level, such as a Pine Script strategy, put the Autoview command in the strategy's alert message and use the webhook path the same way.

Where to go next

If you want the technical walkthrough, the MCP getting-started guide covers the entry, events, and alert tools, token setup, and a full first order end to end. If what you actually want is a TradingView alert executing without any AI in the loop, that's how to automate TradingView alerts, the path Autoview has run at scale for years. For the product overview rather than a how-to, /mcp/ covers what the MCP is in a few sentences. If you're comparing where AI agents fit against Autoview's other surfaces (browser, cloud, webhook), Platforms lays out all four side by side.

Get your MCP token