analyze data within the portfolio - with granular control of what is shared with AI (eg. only %, only currently visible data, filtered data, everything, etc.)
research the web and available market data - in the context of portfolio
navigate within portfolio
modify portfolio - agentic AI
present charts and table within the chat (with option to further explore)
import transactions from non-standard sources, like PDFs, screenshots
daily task to research the need to hedge my portfolio (prompt gathers ~30 metrics on the overall market and reports if they have changed significantly)
weekly task to evaluate portfolio holdings against macro market conditions, check if active investments are still a good play what are new plays
ad hoc analysis (loads of documents from my favorite investors in context), evaluate portfolio + what asset class or individual asset not correlated to my portfolio would improve it
Capitally MCP to provide context to any LLM would be great but it probably clashes with security model of the service.
Yep, all calculations are done on device, so MCP is out of the question.
What could be done though, is to upload the same content metrics we would expose to AI internally, under a known URL with an access token. It would have to be updated from the app, but it could be automated - eg uploaded once a day when you open the app. And for that, you can have MCP.