TradingView MCP Server: What an AI Assistant Can Access, Change and Not Verify

TradingView's MCP server, in public beta since 16 September 2026, lets an AI assistant such as Claude or ChatGPT work with your TradingView account. Through it, the assistant can read quotes, price history, screener results, news, filings and calendars, and it can create, change and delete your watchlists and alerts. The documented toolset contains no tool for placing orders.

For anyone who automates with alerts, the useful line is where the assistant's reach ends. An assistant can set up a simple price alert, even one with a message and a webhook URL, and it can read TradingView's alert log afterwards. Whether that message was a valid instruction, whether PineConnector received and processed it, and what the broker did are separate checks that this server does not cover.

What changed

TradingView announced the MCP server on 16 September 2026 (the post was updated on 21 September 2026)^[1]. MCP, the Model Context Protocol, is an open standard for connecting AI assistants to outside tools and data. The server gives any compatible assistant an official, documented way to call TradingView's tools from a conversation, once you add it as a connector and sign in.

The main facts from the announcement and TradingView's MCP documentation^[2], as of 23 September 2026:

• Who can use it: "Essential or higher-tier plans". The documentation adds that "trial plans don't include MCP access".
• How it connects: one server URL, https://mcp.tradingview.com/mcp, added as a custom connector in Claude, ChatGPT, Codex or another client that supports MCP over streamable HTTP. Sign-in uses OAuth 2.1 with your TradingView account, so there are no API keys to manage.
• Beta limits: TradingView says the beta has a limited tool list and delayed market data, that some tools may not work as smoothly as expected, and that it may limit the number of requests per day^[1]. The documentation states that tool calls are rate-limited to about 100 requests per minute per user.

What the assistant can read and change

The documentation lists 35 tools in nine groups and labels each one Read-only or Write. By TradingView's definition, Read-only tools do not modify the account and Write tools create, change or delete user data^[2].

• Watchlists (8): list, get and get active watchlist are labelled Read-only; create, delete, add symbols, remove symbols and rename or describe are Write.
• Market data (3), symbol search (1), screener (5), news (2), fundamentals and forecasts (3), documents (2), calendars (3): all Read-only. They cover price bars, economic data, symbol lookup, screens and technical ratings, headlines and stories, financials and forecasts, filings and transcripts, and earnings, economic and dividend calendars.
• Alerts (8): list, get details and read the alert log are Read-only; create, update, delete, stop and restart are Write.

Ten tools are labelled Write, and all ten change watchlists or alerts. One tool labelled Read-only can change your account as well. The documentation says that get_active_watchlist activates your first custom watchlist if none is active, and creates a default watchlist if you have none at all^[2]. Treat the label as a guide, not as proof that nothing changes. None of the 35 tools places, modifies or closes a trade. The toolset is described as a beta that expands over time, so check the current list before relying on this summary.

The alert tools in detail

The alert tools matter most to anyone whose alerts drive automation, because their defaults and limits differ from what you might expect in the TradingView dialog^[2].

• Creating an alert. The creation tool makes "a TradingView alert with a simple price condition". The condition types are cross, cross up, cross down, greater and less. The message can be up to 4000 characters and the name up to 300.
• Its defaults. Unless the assistant sets them otherwise, the alert is evaluated on a 1-minute resolution and expires 30 days after creation. It also stays active after it fires, because the option to deactivate it after its first fire defaults to off. Mobile push and pop-up notifications are on by default; email is off.
• Webhooks. The tool accepts a webhook URL, which it says "requires TV plan with webhooks and 2FA". There is also an opt-in "monitoring webhook for LLM analysis on fire", which the documentation does not describe further.
• Changing an alert. The update tool can change the message, name, expiry, notifications and webhook, and it reactivates the alert. "Conditions/symbol/timeframe can NOT be changed". The documented route is to delete and create instead.
• Reading alerts back. The details tool filters out the message and the webhook URL and returns a has_webhook flag instead. The alert log tool returns when each alert fired, with its symbol, message and webhook delivery status, looking back seven days by default.
• Deleting. Deleting removes the alerts and their fire history. Stopping pauses an alert and keeps its history and settings.

One mismatch is worth knowing. The announcement's sample prompts include an alert "when the 1-hour RSI drops below 30"^[1], while the tool reference on the same date describes alert creation as a simple price condition only. We have not tested which applies. Treat the documented tool reference as the narrower, safer description, and check what was actually created.

Where research and alert management stop

The alert log's webhook delivery status is the last point the assistant can see. It tells you what TradingView reported about sending the request, not what happened next. Every later stage keeps its evidence somewhere else, and PineConnector's own guidance keeps them apart: "A received alert is not proof of a broker fill"^[3].

Read the path in order. The first five checkpoints are ones the assistant can take part in; the last four have their own evidence outside the toolset.

1. Research answer. Built on data that is delayed during the beta. Check live prices yourself before any decision.
2. Watchlist created or edited. The list exists in your account. A watchlist sends nothing by itself.
3. Price alert created. An alert object with a supported condition and the tool's defaults. Open it in TradingView and check the symbol, condition, resolution, expiry, trigger frequency, notifications and any notification schedule.
4. Message and webhook URL added. TradingView will send that text to that address when the alert fires. Check every field against the PineConnector Syntax reference, then run the demo test.
5. Webhook delivery status in the log. TradingView's record of the delivery attempt, and the last point the assistant can see.
6. Message received and processed. Evidence: the signal's entry and processing result in PineConnector's Bridge.
7. Order accepted or rejected. Evidence: the terminal's Journal and the broker's response. MetaTrader's help describes an order being checked on the trade server and then executed, cancelled or rejected^[10], so an accepted request is not yet a trade.
8. Deal filled. Evidence: a deal in the account's history, with its price and volume. In MetaTrader a deal is the result of an order being executed^[10].
9. Position open with its stop. Evidence: the position in the Trade tab, with its volume and stop loss. A filled deal opens, adds to, reduces or closes a position^[10].

Two defaults deserve a second look if an assistant ever creates an alert that carries an instruction. An alert that stays active after firing can fire again whenever its condition is met, until it expires or TradingView stops it. The MCP reference does not document its trigger frequency, so check that in TradingView^[2]. The 30-day default expiry means the alert later stops without any other change. Neither default is wrong for a notification. Both change what an automated setup receives.

Your alert messages are visible to the assistant

The alert log tool returns the text of messages that fired. A PineConnector alert message starts with your License ID and, if you use Signal Authentication, includes your secret= value. When an assistant reads that log, those values become part of the conversation.

TradingView already warns against putting credentials or passwords in a webhook body^[4]. PineConnector's Signal Authentication guide asks you to keep the key out of screenshots, support messages and pages copied into an AI assistant, and to replace a key that was shared by accident rather than keep using it^[5]. The same care applies to a connected assistant. Decide whether you want it reading the log of alerts that carry trading instructions, and treat a transcript that includes a secret as a reason to rotate that secret.

Related AI features

Two earlier TradingView releases overlap with the MCP server. The AI Chart Copilot, a browser extension in public beta since 2 April 2026, can set alerts at levels it identifies and can list, pause, restart and delete alerts through conversation^[6]. The AI Screener, in public beta since 17 August 2026, builds Stock Screener filters from a sentence, and TradingView describes its results as a starting point for your own research, not financial advice^[7]. Neither announcement describes placing trades. The same distinction between managing an alert and verifying a trade applies to both.

Limitations and common mistakes

• Treating beta data as live data. TradingView states that market data is delayed during the beta. Do not use the assistant's price readings to time an entry.
• Assuming a prompt equals the result. An assistant may describe an alert it could not create in that form. Open the alert and read what exists.
• Editing conditions in place. Conditions cannot be updated. If an assistant recreates an alert with new conditions, check that the old one was deleted; otherwise both remain active and either can trigger. The same risk applies to TradingView alerts generally, as explained in why an alert keeps old settings.
• Forgetting that write tools act on the real account. Deleting an alert through the assistant also deletes its fire history.
• Reading "delivered" as "traded". Webhook delivery status ends at TradingView's side of the request. An accepted order, a filled deal and an open position are three further checks at the broker.

A sensible division of work

The documented toolset suits research and housekeeping well: finding which alerts fired this week, spotting alerts that carry a webhook through the has_webhook flag, clearing out stale watchlists, and reading filings. For alerts that send instructions to PineConnector, keep creation and changes under human review:

1. If an assistant drafts a PineConnector message, check the License ID, command, exact broker symbol and explicit units yourself against the syntax reference^[8].
2. Open the resulting alert in TradingView and confirm its condition, resolution, trigger frequency, expiry, notification schedule and webhook URL.
3. Run PineConnector's demo test workflow on a demo account^[9]. Do not point an assistant-created alert at a live account.
4. Compare the same event in three places: TradingView's Alerts log, PineConnector's Bridge, and the broker account.

The division of work also runs the other way. An assistant can help with research wherever it runs, while the terminal that receives PineConnector's instructions needs to stay online. PineConnector Edge hosts MT5 with the PineConnector EA so that terminal does not depend on your own computer; the Edge page explains the current plans, and this guide to AI research and hosted MT5 covers the split in more depth.

Reviewed 23 September 2026. Facts were checked against the linked sources on that date. No MCP connection was made for this article; tool behaviour is described as TradingView documents it, and the toolset is in beta.

Related reading

• TradingView updates for alert-based automation: a practical index
• Use AI for strategy research, let someone else host the terminal
• Can ChatGPT or Claude build a TradingView-to-MT5 bridge?
• How to review an AI-generated webhook-to-MT5 project
• TradingView webhook received, but no MT5 trade: a stage-by-stage diagnosis

Sources

1. TradingView Blog – Bring TradingView to Claude with the new MCP Server: public beta available for paid plans, published 16 September 2026, updated 21 September 2026, accessed 23 September 2026.
2. TradingView – MCP Server documentation (setup and tools reference), accessed 23 September 2026.
3. PineConnector Docs – Connect your strategy, accessed 23 September 2026.
4. TradingView Help Center – How to configure webhook alerts, accessed 23 September 2026.
5. PineConnector Docs – Protect signals with a secret key, accessed 23 September 2026.
6. TradingView Blog – TradingView AI Chart Copilot: public beta is available now, published 2 April 2026, updated 19 June 2026, accessed 23 September 2026.
7. TradingView Blog – AI Screener: transform your ideas into ready-made screens, published 17 August 2026, updated 21 August 2026, accessed 23 September 2026.
8. PineConnector Docs – PineConnector Syntax, accessed 23 September 2026.
9. PineConnector Docs – Test your setup, accessed 23 September 2026.
10. MetaTrader 5 Help – Trading operations: basic principles, accessed 23 September 2026.

PineConnector executes the instructions you send it. It does not select trades, manage money, or hold funds. Trading carries risk, and past performance of any strategy does not indicate future results.