A Tech Demo Shakes the Industry
On April 1, 2024, a trader and developer named @Tradesdontlie posted a video that quickly garnered widespread attention in the tech and finance communities. The video demonstrated using a terminal command-line interface to direct Anthropic’s Claude model with natural language commands, controlling the desktop client of the professional charting software TradingView in real-time. Throughout the process, without any manual mouse or keyboard input from the user, the AI autonomously completed a series of complex tasks, including switching trading instruments, writing and compiling Pine Script indicators, and drawing key price levels on the chart. Within 24 hours, the demo attracted over 850,000 views and nearly 5,000 likes, with numerous comments expressing concern about the future of a trader’s career.
Unveiling the Technical Implementation: The “Backdoor” Originates from the Chrome DevTools Protocol
This remarkable functionality doesn’t stem from an official TradingView API but cleverly leverages the technical architecture of its desktop application. The TradingView desktop version is built on the Electron framework, which essentially bundles a Chromium browser core into a standalone application. Chromium has a powerful built-in debugging tool—the Chrome DevTools Protocol (CDP).
CDP is typically used for debugging and performance analysis during development and is disabled by default. However, by appending a specific flag when launching the application, the protocol can be activated, opening an interface that allows external programs to communicate with it. Developer @Axel_bitblaze69 explained the entire process in a detailed technical analysis thread:
- Activate CDP: Modify the launch command for the TradingView desktop app to enable its built-in CDP server port.
- Deploy an MCP Server: Run a local server that implements Anthropic’s Model-Controlled Protocol (MCP).
- Establish the Link: The Claude model sends commands to TradingView’s CDP port via the local MCP server, enabling it to read and modify the application’s DOM (Document Object Model) and simulate user actions.
The entire chain is: Claude Model -> Local MCP Server -> CDP Interface -> TradingView Desktop App. A key advantage of this method is that all data processing and communication occur on the user’s local machine (localhost), without passing through TradingView’s servers, thus posing no risk of data leakage.
The Capabilities of an AI Agent: From Reading Data to Automating Workflows
Through the CDP interface, the Claude model gains unprecedented access, enabling it to “see” and understand almost all information on a TradingView chart, including:
- Basic Data: The current trading instrument, time frame, and the Open (O), High (H), Low (L), Close ©, and Volume (V) of up to 500 candlesticks.
- Indicators & Drawings: The names, IDs, and output values of all indicators on the chart, as well as user-drawn objects like trend lines, rectangles, and text annotations.
- In-depth Information: Results from the Strategy Tester, trade lists, order book depth, and even the rendered values and lines of protected, paid indicators on the chart.
This deep contextual awareness completely transforms the experience of AI-assisted Pine Script coding. Previously, developers had to repeatedly copy and paste code between an AI chat window and the Pine Editor, debugging manually. Now, the process is restructured into an automated closed loop:
- Describe Requirement: The user outlines the strategy requirement in natural language.
- Generate & Inject Code: Claude writes the Pine Script code and automatically injects it into the TradingView editor.
- Compile & Debug: The AI triggers compilation, reads the returned error messages, and autonomously modifies the code based on the errors.
- Iterate & Optimize: Step 3 is repeated until the code compiles successfully and is loaded onto the chart.
This revolution elevates the role of AI from a “code snippet generator” to an “automated engineer” capable of executing complete development and validation workflows.
From Demo to Practice: Real-World Application Within Two Days
The technology’s impact quickly spread from a proof-of-concept demo to practical application. Just two days after the demo’s release, on April 3, user @danielrucci reported his own practical use case. He first had Claude build a backtesting engine and used two years of market data to test a strategy across 17 different trading instruments. After identifying a strategy that passed forward-testing, he used the method published by @Tradesdontlie to connect Claude directly to his TradingView chart, preparing to monitor the strategy’s performance in a live market environment.
This rapid real-world application demonstrated the technology’s practicality and feasibility, intensifying community discussions about AI replacing human traders. Crypto opinion leaders like @Eljaboom went as far as posting headlines like “RIP TRADERS,” further amplifying the event’s hype.
A Sober Look: Risks and Broader Horizons
Amid the heated discussion, calmer voices have pointed out the current challenges and risks of this technology. First, strategy effectiveness is the core issue. An AI can quickly generate strategies that perform perfectly on historical data, but this can easily fall into the trap of “overfitting,” offering no guarantee of profitability in future markets. Creating genuinely robust trading strategies remains a major challenge in the quantitative finance field.
Second, security and stability cannot be overlooked. Opening the CDP port is equivalent to creating a high-privilege backdoor for the desktop application. If not configured properly (e.g., not restricted to local access), it could lead to severe security vulnerabilities. Furthermore, this unofficial integration method is highly dependent on TradingView’s current internal structure. A software update could potentially break the entire automation workflow.
Despite the risks, this event reveals an irreversible trend: AI Agents are bypassing traditional API limitations to interact deeply and directly with existing professional software. In theory, any application built on Electron, such as Figma, Notion, Slack, and VS Code, could be controlled by AI in a similar fashion. This heralds the dawn of a new era where professional workflows will be reshaped, and a vast number of repetitive, tedious operational steps could be replaced by an AI Agent in a terminal.
