AI Can Access Your Portfolio. Is It Now Time to Trust It as an Adviser?

Investors are using AI for stock buy-sell calls, but rules must evolve to address concerns and ensure safeguards

- Jash Kriplani

Some Indian investors are now asking artificial intelligence (AI) assistants to tell them which stocks to buy and sell—directly from inside their trading accounts. This new frontier in retail investing has been made possible by a quiet but powerful innovation: the Model Context Protocol (MCP).

Zerodha, one of India's largest stockbrokers, has become the first to integrate MCP with Claude AI, a large language model (LLM) built by Anthropic. With a few simple technical steps, users can now prompt the AI assistant to access and analyze their portfolio, delivering insights in plain English.

Want to know your portfolio's Sharpe ratio, diversification score, or unrealized losses? Just ask. Claude can also estimate portfolio volatility, assess F&O positions, or even suggest buy/sell actions based on investor-defined goals or benchmarks—all through conversational prompts.

This could be a prompt that simply asks to 'list my stocks'; current positions; unrealized profit and losses; available margin or buying power; analysis of portfolio's risk across metrics like beta; estimated volatility of portfolio; diversification score; Sharpe ratio; analysis of stock portfolio; analysis of open F&O positions.

That's now possible thanks to MCP, an open-source framework released in November 2024 that allows large language models like Claude to access third-party data and tools. By setting up an MCP server, users can give the AI controlled access to their apps and accounts—effectively creating a prompts-driven, portfolio-aware financial assistant.

Early experiments While Zerodha is the only broker to officially offer MCP integration so far, some users have managed to connect Claude with other platforms like Groww—though it requires more technical skill. And tech-savvy investors have begun sharing their experiments online.