Connect your AI assistant (MCP)
Work your portfolio by talking — your own assistant reads it, records what you tell it, and proposes. It cannot trade.
The app isn't the only way in. MCP — the Model Context Protocol, the open standard AI assistants use to reach outside tools — lets the assistant you already use connect directly to your account and work alongside you.
This is the surface for getting things in. Transcribing a manager's book, anchoring eighty positions, filing a dozen claims from a transcript: all of it is typing, and none of it should be yours to do. You describe what you want; it does the work and reports back; you approve.
The app remains where you look — the dashboard view, the charts, and the place to tweak one thing in context while you're staring at the data.
What it can do for you
- Read your state — holdings, sources and their snapshots, the target, freshness and provenance: the same facts the app shows you.
- Record what you tell it — a manager's new disclosure as a dated snapshot, a correction to one you got wrong, the trades a manager disclosed, a claim for the prediction ledger.
- Do the batch work — propose conviction custodians for your whole book with per-position reasoning, then apply the batch once you've approved it in conversation.
- Raise flags — orphaned positions, stale sources, drift against your target, brought to you where you already are.
What it can never do
The tools to trade, move money, or touch a brokerage account do not exist on this surface. Not as an instruction the assistant is asked to respect — as tools that aren't there to call. It cannot place an order, approve a proposal on your behalf, or change your holdings.
There's a second boundary worth stating, because it's the one that actually matters day to day: the assistant executes decisions you state, it does not infer decisions you haven't. It will happily anchor thirty positions the way you just described. It will not decide which manager deserves credit for a position, or what you believe about a company, and each surface publishes its own list of things it will never do.
Connecting one
Two connectors are available, and they don't overlap — most people want both.
Portfolio Manager — everything about your book and the managers you follow: sources,
snapshots, corrections, the copyable book, custodians, orphans, fidelity, blends and
targets, predictions. Its endpoint ends in /manager/mcp:
https://stocks.dataconnector-pro.com/api/v1/manager/mcp
Market analytics — ticker-level research: fundamentals and estimates, snapshots,
comparisons, discounted-cash-flow scenarios, screening. Its endpoint ends in
/chatgpt/mcp:
https://stocks.dataconnector-pro.com/api/v1/chatgpt/mcp
In your assistant's connector settings, add a custom connector pointing at the full URL. Registration and sign-in are automatic — the connection uses standard OAuth with per-client registration and PKCE, so there's no client id to copy and no API key to paste into a chat window.
Once connected, a good first message is simply asking the assistant to describe the surface. It reports who it's authenticated as and what it can actually do, from the live tool list — so it orients itself instead of guessing.
Connected as you, and logged
Your assistant authenticates as you, with a token scoped to exactly the tool surface you approved and nothing more. There's no shared service account.
Every write it makes is recorded as an activity event you can review in the app, so "what did the AI change while I was asleep?" has a complete answer.